Publication List

dj_papers.bib

DEPRECATED. See https://www.seas.upenn.edu/~dineshj/pubs.bib

@article{watson2026let,
  title     = {Let it Cook: Learning to Wait in Sequential Decision Making},
  author    = {Watson, Christopher and Krishna, Arjun and Jayaraman, Dinesh and Alur, Rajeev},
  journal = {Reinforcement Learning Conference (RLC)},
  year      = {2026},
  url       = {https://openreview.net/forum?id=eA56bCxlFT},
  abstract  = {In sequential decision making, an agent typically observes its environment and acts at every timestep. However, such active participation may not always be necessary; tasks such as brewing coffee include periods that are served equally well by letting the environment evolve without constant monitoring and control. During such periods, the agent could simply wait to conserve its resources, or redirect its attention to another task. We capitalize on these opportunities by training a "waiting policy" that decides where and how long to wait. This involves forgoing sensing to commit to a wait action, representing a deliberate pause for a set number of timesteps. We formalize "learning to wait" as minimizing the frequency of sensing and decision making without sacrificing task performance (e.g., the total amount of time to complete a task). To train a waiting policy, we propose an approach that employs reinforcement learning with lexicographically ordered objectives. In experiments across 4 discrete-state household tasks and 3 continuous-state environments, we show that our approach successfully learns waiting behaviors, and can adapt pre-trained policies to wait where appropriate. While different tasks permit different amounts of waiting without sacrificing task performance, our approach consistently finds solutions with significant waiting, sometimes waiting for up to ~50 percent of the task duration.},
  url_openreview = {https://openreview.net/forum?id=eA56bCxlFT},
}
@article{shen2026tiptop,
  title={TiPToP: A Modular Open-Vocabulary Planning System for Robotic Manipulation},
  author={Shen, William and Kumar, Nishanth and Chintalapudi, Sahit and Wang, Jie and Watson, Christopher and Hu, Edward and Cao, Jing and Jayaraman, Dinesh and Kaelbling, Leslie Pack and Lozano-P{\'e}rez, Tom{\'a}s},
  journal={arXiv preprint arXiv:2603.09971},
  abstract={We present TiPToP, an extensible modular system that combines pretrained vision foundation models with an existing Task and Motion Planner (TAMP) to solve multi-step manipulation tasks directly from input RGB images and natural-language instructions. Our system aims to be simple and easy-to-use: it can be installed and run on a standard DROID setup in under one hour and adapted to new embodiments with minimal effort. We evaluate TiPToP -- which requires zero robot data -- over 28 tabletop manipulation tasks in simulation and the real world and find it matches or outperforms π0.5-DROID, a vision-language-action (VLA) model fine-tuned on 350 hours of embodiment-specific demonstrations. TiPToP's modular architecture enables us to analyze the system's failure modes at the component level. We analyze results from an evaluation of 173 trials and identify directions for improvement. We release TiPToP open-source to further research on modular manipulation systems and tighter integration between learning and planning.},
  url={https://tiptop-robot.github.io/},
  year={2026},
  url_pdf = {/publication/shen-2026-tiptop/shen-2026-tiptop.pdf},
  url_project = {https://tiptop-robot.github.io/},
  url_code = {https://github.com/tiptop-robot/tiptop},
  url_arxiv = {https://arxiv.org/abs/2603.09971},
}
@misc{song2026omniguideuniversalguidancefields,
      title={OmniGuide: Universal Guidance Fields for Enhancing Generalist Robot Policies}, 
      author={Yunzhou Song and Long Le and Yong-Hyun Park and Jie Wang and Junyao Shi and Lingjie Liu and Jiatao Gu and Eric Eaton and Dinesh Jayaraman and Kostas Daniilidis},
      year={2026},
      eprint={2603.10052},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2603.10052}, 
	  abstract={Vision-language-action(VLA) models have shown great promise as generalist policies for a large range of relatively simple tasks. However, they demonstrate limited performance on more complex tasks, such as those requiring complex spatial or semantic understanding, manipulation in clutter, or precise manipulation. We propose OMNIGUIDE, a flexible framework that improves VLA performance on such tasks by leveraging arbitrary sources of guidance, such as 3D foundation models, semantic-reasoning VLMs, and human pose models. We show how many kinds of guidance can be naturally expressed as differentiable energy functions with task-specific attractors and repellers located in 3D space, that influence the sampling of VLA actions. In this way, OMNIGUIDE enables guidance sources with complementary task-relevant strengths to improve a VLA model's performance on challenging tasks. Extensive experiments in both simulation and real-world environments, across diverse sources of guidance, demonstrate that OMNIGUIDE enhances the performance of state-of-the-art generalist policies (e.g., π0.5, GR00T N1.6) significantly across success and safety rates. Critically, our unified framework matches or surpasses the performance of prior methods designed to incorporate specific sources of guidance into VLA policies.},
      url_pdf = {/publication/song-2026-omniguideuniversalguidancefields/song-2026-omniguideuniversalguidancefields.pdf},
      url_project = {https://omniguide.github.io/},
}
@article{liu2026timerewarder,
title={TimeRewarder: Learning Dense Reward from Passive Videos via Frame-wise Temporal Distance},
author={Yuyang Liu and Chuan Wen and Yihang Hu and Dinesh Jayaraman and Yang Gao},
journal={ICML},
year={2026},
url={https://openreview.net/forum?id=XztRm216YS},
note={ICML 2026 Spotlight},
abstract={Designing dense rewards is crucial for reinforcement learning (RL), yet in robotics it often demands extensive manual effort and lacks scalability. One promising solution is to view task progress as a dense reward signal, as it quantifies the degree to which actions advance the system toward task completion over time. We present TimeRewarder, a simple yet effective reward learning method that derives progress estimation signals from passive videos, including robot demonstrations and human videos, by modeling temporal distances between frame pairs. We then demonstrate how TimeRewarder can supply step-wise proxy rewards to guide reinforcement learning. In our comprehensive experiments on ten challenging Meta-World tasks, we show that TimeRewarder dramatically improves RL for sparse-reward tasks, achieving nearly perfect success in 9/10 tasks with only 200,000 interactions per task with the environment. This approach outperforms previous methods and even the manually designed environment dense reward on both the final success rate and sample efficiency. Moreover, we show that TimeRewarder pretraining can exploit real-world human videos, highlighting its potential as a scalable approach to rich reward signals from diverse video sources.},
  url_pdf = {/publication/liu-2026-timerewarder/liu-2026-timerewarder.pdf},
  url_project = {https://timerewarder.github.io/},
  url_openreview = {https://openreview.net/forum?id=XztRm216YS},
  url_arxiv = {https://arxiv.org/pdf/2509.26627},
}
@article{qian2026scaling,
  title     = {Expanding Spatial and Temporal Context for Robotic Imitation Learning Policies With Scene Graphs},
  author    = {Qian, Jianing and Peng, Qinhe and Panov, Emmanuel and Fermoselle, Leonor and Jayaraman, Dinesh* and Bucher, Bernadette* and Kelestemur, Tarik*},
  year      = {2026},
  url       = {https://openreview.net/forum?id=WjVTkXDc2T},
  journal   = {CVPR},
  abstract  = {Imitation learning enables robots to learn how to execute tasks via observation. However, real-world environments like homes and offices are often severely partially observed due to their large spatial scales. In addition, many tasks involve executing a series of subtasks requiring autonomous robots to reason over extended time horizons. To address these challenges, we propose using scene graphs as an explicit and structured memory mechanism in imitation learning. By maintaining a dynamic scene graph that captures object-centric relationships and their evolution over time, our method allows the agent to retain relevant historical context during task execution to efficiently reason over incrementally accrued scene information. Our experiments on simulated mobile manipulation and real-world tabletop manipulation demonstrate that our approach substantially improves policy performance, particularly in settings that demand long-term reasoning and robust generalization under partial observability.},
  url_pdf = {/publication/qian-2026-scaling/qian-2026-scaling.pdf},
  url_project = {https://sites.google.com/view/objgraph},
  url_openreview = {https://openreview.net/forum?id=WjVTkXDc2T},
}
@article{shi2026maestro,
      title={Maestro: Orchestrating Robotics Modules with Vision-Language Models for Zero-Shot Generalist Robots}, 
      author={Junyao Shi and Rujia Yang and Kaitian Chao and Selina Bingqing Wan and Yifei Shao and Jiahui Lei and Jianing Qian and Long Le and Pratik Chaudhari and Kostas Daniilidis and Chuan Wen and Dinesh Jayaraman},
      year={2026},
      abstract={Today's best-explored routes towards generalist robots center on collecting ever larger "observations-in actions-out" robotics datasets to train large end-to-end models, copying a recipe that has worked for vision-language models (VLMs). We pursue a road less traveled: building generalist policies directly around VLMs by augmenting their general capabilities with specific robot capabilities encapsulated in a carefully curated set of perception, planning, and control modules. In Maestro, a VLM coding agent dynamically composes these modules into a programmatic policy for the current task and scenario. Maestro's architecture benefits from a streamlined closed-loop interface without many manually imposed structural constraints, and a comprehensive and diverse tool repertoire. As a result, it largely surpasses today's VLA models for zero-shot performance on challenging manipulation skills. Further, Maestro is easily extensible to incorporate new modules, easily editable to suit new embodiments such as a quadruped-mounted arm, and even easily adapts from minimal real-world experiences through local code edits.},
	  journal={(Under review)},
      url_pdf = {/publication/shi-2026-maestro/shi-2026-maestro.pdf},
      url_project = {https://maestro-robot.github.io/},
      url = {https://maestro-robot.github.io/static/Maestro.pdf},
}
@article{yang2026m3ocap,
  title={Motion Capture with Millimeter-Wave Tags},
  author={Xin Yang and Freddy Yifei Liu and Yunshuang Li and Yao Gong and Dinesh Jayaraman and Omid Abari and Mingmin Zhao},
  journal={SenSys},
  year={2026},
  url_pdf = {/publication/yang-2026-m-3-ocap/yang-2026-m-3-ocap.pdf},
}
@article{le2025pixie,
  title={Pixie: Fast and Generalizable Supervised Learning of 3D Physics from Pixels},
  author={Le, Long and Lucas, Ryan and Wang, Chen and Chen, Chuhao and Jayaraman, Dinesh and Eaton, Eric and Liu, Lingjie},
  journal={arXiv preprint arXiv:2508.17437},
  year={2025},
  url_pdf = {/publication/le-2025-pixie/le-2025-pixie.pdf},
  url_project = {https://pixie-3d.github.io/},
  url_code = {https://github.com/vlongle/pixie},
  url = {https://arxiv.org/abs/2508.17437},
}
@article{shipoints2reward,
  title={Points2Reward: Robotic Manipulation Rewards from Just One Video},
  author={Shi, Junyao and Smith, Joshua and Qian, Jianing and Jayaraman, Dinesh},
  journal={(Under review)},
  year={2025},
  abstract={Detailed and dense reward functions provided clearly graded signals that enable a robot to evaluate and improve its policy, but generating such functions for new tasks is often cumbersome and expertise-intensive. On the other hand, end users can often easily record a video demonstrating the desired object trajectories and goal configurations involved in a new skill, but we lack methods to reliably learn from such a simple specification. We propose “Points2Reward” (P2R), which effectively computes dense rewards from a single video. To do this, P2R tracks task-relevant object points in task demonstrations and policy rollouts, and matches them to compare the desired and achieved object trajectories to generate reward scores. Exploiting recent advances in point tracking and semantic point correspondences, P2R produces high-quality rewards even under significant domain gaps between the demo video and the robot setup, such as embodiment gaps (human vs. robot) or camera viewpoint changes. We demonstrate that P2R correctly evaluates trajectories of varying quality in diverse real-world settings as well as in nine simulated manipulation tasks in standard benchmark suites. We further demonstrate that P2R policy evaluations enable improved downstream policy synthesis in the simulated tasks.},
  url_pdf = {/publication/shipoints-2-reward/shipoints-2-reward.pdf},
  url_project = {https://points2reward.github.io/},
}
@article{gao2026vlmgineer,
  title={VLMgineer: Vision-Language Models as Robotic Toolsmiths},
  author={Gao, George Jiayuan and Li, Tianyu and Shi, Junyao and Li, Yihan and Zhang, Zizhe and Figueroa, Nadia and Jayaraman, Dinesh},
  journal={ICLR},
  year={2026},
  abstract={Tool design and use reflect the ability to understand and manipulate the physical world through creativity, planning, and foresight. As such, it is often regarded as a measurable indicator of cognitive intelligence across biological species. While much of today's research on robotics intelligence focuses on generating better control strategies, inventing smarter tools offers a complementary form of physical intelligence: moving the problem-solving onus into the tool's geometry so that control becomes simpler. This motivates us to ask: can today's foundation models offer useful priors to automatically invent—and effectively wield—such tools? We present VLMgineer, a framework that harnesses the creativity of Vision–Language Models (VLMs) together with evolutionary search to co-design physical tools and the control policies that operate them. We evaluate VLMgineer on a diverse benchmark of everyday manipulation scenarios that demand creative tool design and use. Across this suite, VLMgineer consistently discovers tools and policies that solve tasks more effectively and innovatively, transforming challenging robotics problems into straightforward executions. It also consistently outperforms VLM-generated designs from human specifications and existing human-crafted tools for everyday tasks. To facilitate future research on automated tool invention, we will release our benchmark and code.},
  url_pdf = {/publication/gao-2026-vlmgineer/gao-2026-vlmgineer.pdf},
  url_project = {https://vlmgineer.github.io/release},
  url = {https://arxiv.org/abs/2507.12644},
}

@article{liang2026trajectory,
  title={Correspondence-Driven Trajectory Warping for Data-Efficient Imitation and Autonomous Play},
  author={Liang, William and Wang, Sam and Wang, Hung-Ju and Bastani, Osbert and Ma, Yecheng Jason and Jayaraman, Dinesh},
  journal={ICLR},
  year={2026},
  abstract={Imitation learning has emerged as a promising paradigm for learning robotic manipulation policies, but leading methods require large, expensive datasets of human-collected demonstrations. Towards data-efficient imitation learning, we propose a novel non-parametric policy that produces useful behaviors from a few demonstrations. Given two-view images, it identifies semantic correspondences to anchor warps of demonstration trajectories into new real-world scenes. We show that this design is efficient and robust to significant variations in the spatial and semantic configuration of the scene, such as dramatic positional differences and out-of-distribution objects. As a result, the policy excels at a variety of manipulation tasks involving deformation, complex contacts, articulation, and precision, highlighting its flexibility and generality. In addition, we demonstrate how a bank of such policies can power autonomous multi-task play in the real world via a continuous cycle of task selection, execution, evaluation, and policy improvement, guided by vision-language models. This procedure generates an increasingly diverse demonstration dataset over time for each task, while minimizing the need for manual resets and human interventions. In a real household-like multi-object environment, our method is among the first to bootstrap many hours of diverse play from few demonstrations, autonomously producing hundreds of expert-level trajectories that could be used for downstream policy learning.},
  url_pdf = {/publication/liang-2026-trajectory/liang-2026-trajectory.pdf},
  url_project = {https://tether-research.github.io},
  url_code = {https://github.com/tether-research/tether},
  url_openreview = {https://openreview.net/forum?id=FqDmvMZish},
  url = {https://arxiv.org/abs/2603.03278},
}

@article{hu2025interactive,
  title={Real-World Reinforcement Learning of Interactive Perception Behaviors},
  author={Hu, Edward S. and Wang, Jie and Yuan, Xingfang and Luo, Fiona and Li, Muyao and Lambrechts, Gaspard and Rybkin, Oleh and Jayaraman, Dinesh},
  journal={NeurIPS},
  year={2025},
  abstract={A robot's instantaneous sensory observations do not always reveal all task-relevant state information. Under such partial observability, optimal behavior typically involves explicitly acting to gain the missing information. Today's standard robot learning techniques struggle to produce such active perception behaviors. We propose a simple real-world robot learning recipe to efficiently train active perception policies. Our approach, asymmetric advantage weighted regression (AAWR), exploits access to "privileged" extra sensors at training time. The privileged sensors enable training high-quality privileged value functions that aid in estimating the advantage of the target policy. Bootstrapping from a small number of potentially suboptimal demonstrations and an easy-to-obtain coarse policy initialization, AAWR quickly acquires active perception behaviors and boosts task performance. In evaluations on 8 manipulation tasks on 3 robots spanning varying degrees of partial observability, AAWR synthesizes reliable active perception behaviors that outperform all prior approaches. When initialized with a "generalist" robot policy that struggles with active perception tasks, AAWR efficiently generates information-gathering behaviors that allow it to operate under severe partial observability for manipulation tasks.},
  url_pdf = {/publication/hu-2025-interactive/hu-2025-interactive.pdf},
  url_project = {https://penn-pal-lab.github.io/aawr/},
  url_code = {https://github.com/penn-pal-lab/aawr},
  url_video = {https://www.youtube.com/watch?v=4yzFTFli-wc},
  url_openreview = {https://openreview.net/forum?id=RkdTtznSAL},
  url = {https://arxiv.org/abs/2512.01188},
}

@article{sridhar2025ricl,
title={RICL: Adding In-Context Adaptability to Pre-Trained Vision-Language-Action Models},
author={Kaustubh Sridhar and Souradeep Dutta and Dinesh Jayaraman and Insup Lee},
abstract={Multi-task ``vision-language-action'' (VLA) models have recently demonstrated increasing promise as generalist foundation models for robotics, achieving non-trivial performance out of the box on new tasks in new environments. However, for such models to be truly useful, an end user must have easy means to teach them to improve. For language and vision models, the emergent ability to perform in-context learning (ICL) has proven to be a versatile and highly useful interface to easily teach new tasks with no parameter finetuning. Unfortunately, VLAs pre-trained with imitation learning objectives do not naturally acquire ICL abilities. In this paper, we demonstrate that, with the right finetuning recipe and a small robot demonstration dataset, it is possible to inject in-context adaptability post hoc into such a VLA. After retraining for in-context learning (RICL), our system permits an end user to provide a small number (10-20) of demonstrations for a new task. RICL then fetches the most relevant portions of those demonstrations into the VLA context to exploit ICL, performing the new task and boosting task performance. We apply RICL to inject ICL into the pi0-FAST VLA, and show that it permits large in-context improvements for a variety of new manipulation tasks with only 20 demonstrations per task, without any parameter updates. When parameter updates on the target task demonstrations is possible, RICL finetuning further boosts performance. We release code and model weights for RICL-pi0-FAST alongside the paper to enable, for the first time, a simple in-context learning interface for new manipulation tasks},
journal={CORL},
year={2025},
  url_pdf = {/publication/sridhar-2025-ricl/sridhar-2025-ricl.pdf},
  url_project = {https://ricl-vla.github.io/},
  url_code = {https://github.com/ricl-vla/ricl_openpi},
  url_video = {https://www.youtube.com/watch?v=ZC6-1DUSNy4},
  url_openreview = {https://openreview.net/forum?id=6AASPlloSt},
  url = {https://arxiv.org/abs/2508.02062},
}

@article{atreya2025roboarena,
	title = {{RoboArena}: {Distributed} {Real}-{World} {Evaluation} of {Generalist} {Robot} {Policies}},
	shorttitle = {{RoboArena}},
	url = {http://arxiv.org/abs/2506.18123},
	doi = {10.48550/arXiv.2506.18123},
	abstract = {Comprehensive, unbiased, and comparable evaluation of modern generalist policies is uniquely challenging: existing approaches for robot benchmarking typically rely on heavy standardization, either by specifying fixed evaluation tasks and environments, or by hosting centralized ''robot challenges'', and do not readily scale to evaluating generalist policies across a broad range of tasks and environments. In this work, we propose RoboArena, a new approach for scalable evaluation of generalist robot policies in the real world. Instead of standardizing evaluations around fixed tasks, environments, or locations, we propose to crowd-source evaluations across a distributed network of evaluators. Importantly, evaluators can freely choose the tasks and environments they evaluate on, enabling easy scaling of diversity, but they are required to perform double-blind evaluations over pairs of policies. Then, by aggregating preference feedback from pairwise comparisons across diverse tasks and environments, we can derive a ranking of policies. We instantiate our approach across a network of evaluators at seven academic institutions using the DROID robot platform. Through more than 600 pairwise real-robot evaluation episodes across seven generalist policies, we demonstrate that our crowd-sourced approach can more accurately rank the performance of existing generalist policies than conventional, centralized evaluation approaches, while being more scalable, resilient, and trustworthy. We open our evaluation network to the community and hope that it can enable more accessible comparisons of generalist robot policies.},
	urldate = {2025-09-23},
	journal = {CORL},
	author = {Atreya, Pranav and Pertsch, Karl and Lee, Tony and Kim, Moo Jin and Jain, Arhan and Kuramshin, Artur and Eppner, Clemens and Neary, Cyrus and Hu, Edward and Ramos, Fabio and Tremblay, Jonathan and Arora, Kanav and Ellis, Kirsty and Macesanu, Luca and Leonard, Matthew and Cho, Meedeum and Aslan, Ozgur and Dass, Shivin and Wang, Jie and Yuan, Xingfang and Yang, Xuning and Gupta, Abhishek and Jayaraman, Dinesh and Berseth, Glen and Daniilidis, Kostas and Martin-Martin, Roberto and Lee, Youngwoon and Liang, Percy and Finn, Chelsea and Levine, Sergey},
	month = jun,
	year = {2025},
	note = {arXiv:2506.18123 [cs]},
	keywords = {Computer Science - Machine Learning, Computer Science - Robotics},
	annote = {Comment: Website: https://robo-arena.github.io/},
	file = {Preprint PDF:/Users/dineshj/Zotero/storage/8AGWFMBI/Atreya et al. - 2025 - RoboArena Distributed Real-World Evaluation of Generalist Robot Policies.pdf:application/pdf;Snapshot:/Users/dineshj/Zotero/storage/LQJH37S2/2506.html:text/html},
	url_pdf = {/publication/atreya-2025-roboarena/atreya-2025-roboarena.pdf},
	url_project = {https://robo-arena.github.io/},
	url_code = {https://github.com/robo-arena/roboarena},
	url_dataset = {https://huggingface.co/datasets/RoboArena/DataDump_02-03-2026},
}


@article{kunapuli2025leveling,
title={Leveling the Playing Field: Carefully Comparing Classical and Learned Controllers for Quadrotor Trajectory Tracking},
author={Pratik Kunapuli and Jake Welde and Dinesh Jayaraman and Vijay Kumar},
abstract={Learning-based control approaches like reinforcement learning (RL) have recently produced a slew of impressive results for tasks like quadrotor trajectory tracking and drone racing. Naturally, it is common to demonstrate the advantages of these new controllers against established methods like analytical controllers. We observe, however, that reliably comparing the performance of these very different classes of controllers is more complicated than might appear at first sight. As a case study, we take up the problem of agile tracking of an end-effector for a quadrotor with a fixed-arm. We develop a set of best practices for synthesizing the best RL and Geometric controllers for benchmarking. In the process, we fix widely prevalent RL-favoring biases in prior studies that provide asymmetric access to: (1) the task definition in the form of objective functions, (2) datasets for parameter optimization, and (3) “feed-forward” controller inputs revealing the desired future trajectory. The resulting contributions are threefold: first, our improved robust experimental protocol reveals that the gaps between the two controller classes are much smaller than expected from previously published findings. Geometric control performs on par or better than RL in most practical settings, while RL fares better in transient performance at the expense of steady-state error. Second, our improvements to the experimental protocol for comparing learned and classical controller synthesis approaches are critical: each of the above asymmetries can yield misleading conclusions, and we show evidence that suggests that they indeed have in prior quadrotor studies. Finally, we open-source implementations of Geometric and RL controllers for these aerial vehicles implementing best practices for future development.},
journal={RSS},
year={2025},
  url_pdf = {/publication/kunapuli-2025-leveling/kunapuli-2025-leveling.pdf},
  url_project = {https://pratikkunapuli.github.io/rl-vs-gc/},
  url_code = {https://github.com/PratikKunapuli/rl-vs-gc},
  url = {https://arxiv.org/abs/2506.17832},
}

@article{watson2025ilg,
title={Illustrated Landmark Graphs for Long-Horizons Policy Learning},
author={Christopher Watson and Arjun Krishna and Rajeev Alur and Dinesh Jayaraman},
abstract={Applying learning-based approaches to long-horizon sequential decision-making tasks requires a human teacher to carefully craft reward functions or curate demonstrations to elicit desired behaviors. To simplify this, we first introduce an alternative form of task-specification, Illustrated Landmark Graph (ILG), that represents the task as a directed-acyclic graph where each vertex corresponds to a region of the state space (a landmark), and each edge represents an easier to achieve sub-task. A landmark in the ILG is conveyed to the agent through a few illustrative examples grounded in the agent’s observation space. Second, we propose ILG-Learn, a human in the loop algorithm that interleaves planning over the ILG and sub-task policy learning. ILG-Learn adaptively plans through the ILG by relying on the human teacher’s feedback to estimate the success rates of learned policies. We conduct experiments on long-horizon block stacking and point maze navigation tasks, and find that our approach achieves considerably higher success rates (~ 50% improvement) compared to hierarchical reinforcement learning and imitation learning baselines. Additionally, we highlight how the flexibility of the ILG specification allows the agent to learn a sequence of sub-tasks that is better suited to its limited capabilities.}, 
journal={TMLR},
year={2025},
  url_pdf = {/publication/watson-2025-ilg/watson-2025-ilg.pdf},
  url_project = {https://ilg-learn.github.io/},
  url_code = {https://github.com/cwatson1998/ilg-learn},
  url_openreview = {https://openreview.net/forum?id=0AOUWC4ss8},
}
@article{krishna2025vosi,
title={The Value of Sensory Information to a Robot},
author={Arjun Krishna and Edward S. Hu and Dinesh Jayaraman},
abstract={A decision-making agent, such as a robot, must observe and react to any new task-relevant information that becomes available from its environment. We seek to study a fundamental scientific question: what value does sensory information hold to an agent at various moments in time during the execution of a task? Towards this, we empirically study agents of varying architectures, generated with varying policy synthesis approaches (imitation, RL, model-based control), on diverse robotics tasks. For each robotic agent, we characterize its regret in terms of performance degradation when state observations are withheld from it at various task states for varying lengths of time. We find that sensory information is surprisingly rarely task-critical in many commonly studied task setups. Task characteristics such as stochastic dynamics largely dictate the value of sensory information for a well-trained robot; policy architectures such as planning vs. reactive control generate more nuanced second-order effects. Further, sensing efficiency is curiously correlated with task proficiency: in particular, fully trained high-performing agents are more robust to sensor loss than novice agents early in their training. Overall, our findings characterize the tradeoffs between sensory information and task performance in practical sequential decision making tasks, and pave the way towards the design of more resource-efficient decision-making agents.},
journal={ICLR},
year={2025},
  url_pdf = {/publication/krishna-2025-vosi/krishna-2025-vosi.pdf},
  url_project = {https://sites.google.com/view/vosi-robotics/},
  url_openreview = {https://openreview.net/forum?id=ikr5XomWHS},
}
@article{ma2025gvl,
title={Vision Language Models are In-Context Value Learners},
author={Yecheng Jason Ma and Joey Hejna and Chuyuan Fu and Dhruv Shah and Jacky Liang and Zhuo Xu and Sean Kirmani and Peng Xu and Danny Driess and Ted Xiao and Osbert Bastani and Dinesh Jayaraman and Wenhao Yu and Tingnan Zhang and Dorsa Sadigh and Fei Xia},
abstract={Predicting temporal progress from visual trajectories is important for intelligent robots that can learn, adapt, and improve. However, learning such progress estimator, or temporal value function, across different tasks and domains requires both a large amount of diverse data and methods which can scale and generalize. To address these challenges, we present Generative Value Learning (GVL), a universal value function estimator that leverages the world knowledge embedded in vision-language models (VLMs) to predict task progress. Naively asking a VLM to predict values for a video sequence performs poorly due to the strong temporal correlation between successive frames. Instead, GVL poses value estimation as a temporal ordering problem over shuffled video frames; this seemingly more challenging task encourages VLMs to more fully exploit their underlying semantic and temporal grounding capabilities to differentiate frames based on their perceived task progress, consequently producing significantly better value predictions. Without any robot or task specific training, GVL can in-context zero-shot and few-shot predict effective values for more than 300 distinct real-world tasks across diverse robot platforms, including challenging bimanual manipulation tasks. Furthermore, we demonstrate that GVL permits flexible multi-modal in-context learning via examples from heterogeneous tasks and embodiments, such as human videos. The generality of GVL enables various downstream applications pertinent to visuomotor policy learning, including dataset filtering, success detection, and value-weighted regression -- all without any model training or finetuning.},
journal={ICLR},
year={2025},
  url_pdf = {/publication/ma-2025-gvl/ma-2025-gvl.pdf},
  url_project = {https://generative-value-learning.github.io/},
  url = {https://www.arxiv.org/abs/2411.04549},
}
@article{hu2025belief,
title={Learning to Achieve Goals with Belief State Transformers},
author={Edward S. Hu and Kwangjun Ahn and Qinghua Liu and Haoran Xu and Manan Tomar and Ada Langford and Dinesh Jayaraman and Alex Lamb and John Langford},
abstract={We introduce the "Belief State Transformer", a next-token predictor that takes both a prefix and suffix as inputs, with a novel objective of predicting both the next token for the prefix and the previous token for the suffix. The Belief State Transformer effectively learns to solve challenging problems that conventional forward-only transformers struggle with, in a domain-independent fashion. Key to this success is learning a compact belief state that captures all relevant information necessary for accurate predictions. Empirical ablations show that each component of the model is essential in difficult scenarios where standard Transformers fall short. For the task of story writing with known prefixes and suffixes, our approach outperforms the Fill-in-the-Middle method for reaching known goals and demonstrates improved performance even when the goals are unknown.
Altogether, the Belief State Transformer enables more efficient goal-conditioned decoding, better test-time inference, and high-quality text representations on small scale problems.},
journal={ICLR},
year={2025},
  url_pdf = {/publication/hu-2025-belief/hu-2025-belief.pdf},
  url_project = {https://sites.google.com/view/belief-state-transformer},
  url_code = {https://github.com/microsoft/BST},
  url_openreview = {https://openreview.net/forum?id=ThRMTCgpvo},
  url = {https://arxiv.org/abs/2410.23506},
}
@article{le2025articulate,
title={Articulate-Anything: Automatic Modeling of Articulated Objects via a Vision-Language Foundation Model},
author={Long Le and Jason Xie and William Liang and Hung-Ju Wang and Yue Yang and Yecheng Jason Ma and Kyle Vedder and Arjun Krishna and Dinesh Jayaraman and Eric Eaton},
abstract={Interactive 3D simulated objects are crucial in AR/VR, animations, and robotics, driving immersive experiences and advanced automation. However, creating these articulated objects requires extensive human effort and expertise, limiting their broader applications. To overcome this challenge, we present Articulate-Anything, a system that automates the articulation of diverse, complex objects from many input modalities, including text, images, and videos. Articulate-Anything leverages vision-language models (VLMs) to generate code that can be compiled into an interactable digital twin for use in standard 3D simulators. Our system exploits existing 3D asset datasets via a mesh retrieval mechanism, along with an actor-critic system that iteratively proposes, evaluates, and refines solutions for articulating the objects, self-correcting errors to achieve a robust out- come. Qualitative evaluations demonstrate Articulate-Anything's capability to articulate complex and even ambiguous object affordances by leveraging rich grounded inputs. In extensive quantitative experiments on the standard PartNet-Mobility dataset, Articulate-Anything substantially outperforms prior work, increasing the success rate from 8.7-11.6% to 75% and setting a new bar for state-of-art performance. We further showcase the utility of our generated assets by using them to train robotic policies for fine-grained manipulation tasks that go beyond basic pick and place.},
journal={ICLR},
year={2025},
  url_pdf = {/publication/le-2025-articulate/le-2025-articulate.pdf},
  url_project = {https://articulate-anything.github.io/},
  url_code = {https://github.com/vlongle/articulate-anything/},
  url = {https://arxiv.org/abs/2410.13882},
}
@article{sridhar2025regent,
title={REGENT: A Retrieval-Augmented Generalist Agent That Can Act In-Context in New Environments},
author={Kaustubh Sridhar and Souradeep Dutta and Dinesh Jayaraman and Insup Lee},
abstract={Do generalist agents require large models pre-trained on massive amounts of data to rapidly adapt to new environments? We propose a novel approach to pre-train relatively small models and adapt them to unseen environments via in-context learning, without any finetuning. Our key idea is that retrieval offers a powerful bias for fast adaptation. Indeed, we demonstrate that even a simple retrieval-based 1-nearest neighbor agent offers a surprisingly strong baseline for today's state-of-the-art generalist agents. From this starting point, we construct a semi-parametric agent, REGENT, that trains a transformer-based policy on sequences of queries and retrieved neighbors. REGENT can generalize to unseen robotics and game-playing environments via retrieval augmentation and in-context learning, achieving this with up to 3x fewer parameters and up to an order-of-magnitude fewer pre-training datapoints, significantly outperforming today's state-of-the-art generalist agents.},
journal={ICLR},
year={2025},
  url_pdf = {/publication/sridhar-2025-regent/sridhar-2025-regent.pdf},
  url_project = {https://kaustubhsridhar.github.io/regent-research/},
  url_code = {https://github.com/regent-research/regent},
  url_openreview = {https://openreview.net/forum?id=NxyfSW6mLK},
  url = {https://arxiv.org/abs/2412.04759},
}
@article{shi2025zeromimic,
title={ZeroMimic: Distilling Robotic Manipulation Skills from Web Videos},
author={Junyao Shi* and Zhuolun Zhao* and Tianyou Wang and Ian Pedroza and Amy Luo and Jie Wang and Jason Ma and Dinesh Jayaraman},
abstract={Many recent advances in robotic manipulation
have come through imitation learning, yet these rely largely
on mimicking a particularly hard-to-acquire form of demon-
strations: those collected on the same robot in the same room
with the same objects as the trained policy must handle at test
time. In contrast, large pre-recorded human video datasets
demonstrating manipulation skills in-the-wild already exist,
which contain valuable information for robots. Is it possible to
distill a repository of useful robotic skill policies out of such
data without any additional requirements on robot-specific
demonstrations or exploration? We present the first such
system ZeroMimic, that generates immediately deployable image
goal-conditioned skill policies for several common categories
of manipulation tasks (opening, closing, pouring, pick&place,
cutting, and stirring) each capable of acting upon diverse objects
and across diverse unseen task setups. ZeroMimic is carefully
designed to exploit recent advances in semantic and geometric
visual understanding of human videos, together with modern
grasp affordance detectors and imitation policy classes. After
training ZeroMimic on the popular EpicKitchens dataset of ego-
centric human videos, we evaluate its out-of-the-box performance
in varied kitchen settings, demonstrating its impressive abilities
to handle these varied tasks. To enable plug-and-play reuse of
ZeroMimic policies on other task setups and robots, we will
release software and policy checkpoints for all skills.},
journal={ICRA},
year={2025},
url={https://zeromimic.github.io/},
  url_pdf = {/publication/shi-2025-zeromimic/shi-2025-zeromimic.pdf},
  url_project = {https://zeromimic.github.io/},
  url_code = {https://github.com/junyaoshi/ZeroMimic},
  url_video = {https://www.youtube.com/watch?v=nFhNSsR7TEY},
  url_arxiv = {https://arxiv.org/abs/2503.23877},
}

@article{welde2025symmetry,
title={Leveraging Symmetry to Accelerate Learning of Trajectory Tracking Controllers for Free-Flying Robotic Systems},
author={Jake Welde* and Nishanth Rao* and Pratik Kunapuli* and Dinesh Jayaraman and Vijay Kumar},
abstract={Tracking controllers enable robotic systems to
accurately follow planned reference trajectories. In particular,
reinforcement learning (RL) has shown promise in the synthesis
of controllers for systems with complex dynamics and modest
online compute budgets. However, the poor sample efficiency of
RL and the challenges of reward design make training slow and
sometimes unstable, especially for high-dimensional systems. In
this work, we leverage the inherent Lie group symmetries of
robotic systems with a floating base to mitigate these chal-
lenges when learning tracking controllers. We model a general
tracking problem as a Markov decision process (MDP) that
captures the evolution of both the physical and reference states.
Next, we prove that symmetry in the underlying dynamics and
running costs leads to an MDP homomorphism, a mapping
that allows a policy trained on a lower-dimensional “quotient”
MDP to be lifted to an optimal tracking controller for the
original system. We compare this symmetry-informed approach
to an unstructured baseline, using Proximal Policy Optimization
(PPO) to learn tracking controllers for three systems: the
Particle (a forced point mass), the Astrobee (a fully-
actuated space robot), and the Quadrotor (an underactuated
system). Results show that a symmetry-aware approach both
accelerates training and reduces tracking error after the same
number of training steps.},
journal={ICRA},
year={2025},
  url_pdf = {/publication/welde-2025-symmetry/welde-2025-symmetry.pdf},
  url_project = {https://pratikkunapuli.github.io/EQTrackingControl/},
  url_code = {https://github.com/PratikKunapuli/EQTrackingControl},
  url_video = {https://www.youtube.com/watch?v=AosGBe2uzxM},
  url = {https://arxiv.org/abs/2409.11238},
}
@article{liang2024eurekaverse,
title = {Environment Curriculum Generation via Large Language Models},
author = {William Liang and Sam Wang and Hungju Wang and
Osbert Bastani and Dinesh Jayaraman* and Yecheng Jason Ma*},
abstract={Recent work has demonstrated that a promising strategy for teaching robots a wide range of complex skills is by training them on a curriculum of progressively more challenging environments. However, developing an effective curriculum of environment distributions currently requires significant expertise, which must be repeated for every new domain. Our key insight is that environments are often naturally represented as code. Thus, we probe whether effective environment curriculum design can be achieved and automated via code generation by large language models (LLM). In this paper, we introduce Eurekaverse, an unsupervised environment design algorithm that uses LLMs to sample progressively more challenging, diverse, and learnable environments for skill training. We validate Eurekaverse's effectiveness in the domain of quadrupedal parkour learning, in which a quadruped robot must traverse through a variety of obstacle courses. The automatic curriculum designed by Eurekaverse enables gradual learning of complex parkour skills in simulation and can successfully transfer to the real-world, outperforming manual training courses designed by humans.},
journal={CORL},
year={2024},
url={https://eureka-research.github.io/eurekaverse/},
  url_pdf = {/publication/liang-2024-eurekaverse/liang-2024-eurekaverse.pdf},
  url_project = {https://eureka-research.github.io/eurekaverse/},
  url_code = {https://github.com/eureka-research/eurekaverse},
  url_video = {https://www.youtube.com/watch?v=kRj3crlFdOU},
  url_arxiv = {https://arxiv.org/abs/2411.01775},
}
@article{qian2024hodor,
  title={Task-Oriented Hierarchical Object Decomposition for Visuomotor Control },
  author = {Jianing Qian and Bernadette Bucher and Dinesh Jayaraman},
  abstract={Good pre-trained visual representations could enable robots to learn visuomotor policy efficiently. Still, existing representations take a one-size-fits-all-tasks approach that comes with two important drawbacks: (1) Being completely task-agnostic, these representations cannot effectively ignore any task-irrelevant information in the scene,  and (2) They often lack the representational capacity to handle unconstrained/complex real-world scenes. Instead, we propose to train a large combinatorial family of representations organized by scene entities: objects and object parts. This \underline{h}ierarchical \underline{o}bject \underline{d}ecomposition for task-\underline{o}riented \underline{r}epresentations (\methodname) permits selectively assembling different representations specific to each task while scaling in representational capacity with the complexity of the scene and the task. In our experiments, we find that \methodname outperforms prior pre-trained representations, both scene vector representations and object-centric representations, for sample-efficient imitation learning across 5 simulated and 5 real-world manipulation tasks. We further find that the invariances captured in \methodname are inherited into downstream policies, which can robustly generalize to out-of-distribution test conditions, permitting zero-shot skill chaining. Appendix and videos: https://sites.google.com/view/hodor-corl24.},
  journal={CORL},
  year={2024},
  url={https://sites.google.com/view/hodor-corl24},
  url_pdf = {/publication/qian-2024-hodor/qian-2024-hodor.pdf},
  url_project = {https://sites.google.com/view/hodor-corl24},
  url_arxiv = {https://arxiv.org/abs/2411.01284},
}
@article{open_x_embodiment_rt_x_2024,
title={Open {X-E}mbodiment: Robotic Learning Datasets and {RT-X} Models},
author = {Large collaboration},
journal  = {ICRA},
year = {2024},
url = {https://robotics-transformer-x.github.io/},
  url_pdf = {/publication/open-x-embodiment-rt-x-2024/open-x-embodiment-rt-x-2024.pdf},
  url_project = {https://robotics-transformer-x.github.io/},
  url_code = {https://github.com/google-deepmind/open_x_embodiment},
  url_arxiv = {https://arxiv.org/abs/2310.08864},
}
@article{ma2024dreureka,
    title   = {DrEureka: Language Model Guided Sim-To-Real Transfer},
    author  = {Yecheng Jason Ma and William Liang and Hungju Wang and Sam Wang and Yuke Zhu and Linxi Fan and Osbert Bastani and Dinesh Jayaraman},
    year    = {2024},
    journal = {RSS},
    url_pdf = {/publication/ma-2024-dreureka/ma-2024-dreureka.pdf},
    url_project = {https://eureka-research.github.io/dr-eureka/},
    url_code = {https://github.com/eureka-research/DrEureka},
    url_video = {https://www.youtube.com/watch?v=eECuJUuSt5c},
    url = {https://arxiv.org/abs/2406.01967},
}
@article{khazatsky2024droid,
      title={DROID: A Large-Scale In-The-Wild Robot Manipulation Dataset}, 
      author={Alexander Khazatsky and Karl Pertsch and Suraj Nair and Ashwin Balakrishna and Sudeep Dasari and Siddharth Karamcheti and Soroush Nasiriany and Mohan Kumar Srirama and Lawrence Yunliang Chen and Kirsty Ellis and Peter David Fagan and Joey Hejna and Masha Itkina and Marion Lepert and Yecheng Jason Ma and Patrick Tree Miller and Jimmy Wu and Suneel Belkhale and Shivin Dass and Huy Ha and Arhan Jain and Abraham Lee and Youngwoon Lee and Marius Memmel and Sungjae Park and Ilija Radosavovic and Kaiyuan Wang and Albert Zhan and Kevin Black and Cheng Chi and Kyle Beltran Hatch and Shan Lin and Jingpei Lu and Jean Mercat and Abdul Rehman and Pannag R Sanketi and Archit Sharma and Cody Simpson and Quan Vuong and Homer Rich Walke and Blake Wulfe and Ted Xiao and Jonathan Heewon Yang and Arefeh Yavary and Tony Z. Zhao and Christopher Agia and Rohan Baijal and Mateo Guaman Castro and Daphne Chen and Qiuyu Chen and Trinity Chung and Jaimyn Drake and Ethan Paul Foster and Jensen Gao and David Antonio Herrera and Minho Heo and Kyle Hsu and Jiaheng Hu and Donovon Jackson and Charlotte Le and Yunshuang Li and Kevin Lin and Roy Lin and Zehan Ma and Abhiram Maddukuri and Suvir Mirchandani and Daniel Morton and Tony Nguyen and Abigail O'Neill and Rosario Scalise and Derick Seale and Victor Son and Stephen Tian and Emi Tran and Andrew E. Wang and Yilin Wu and Annie Xie and Jingyun Yang and Patrick Yin and Yunchu Zhang and Osbert Bastani and Glen Berseth and Jeannette Bohg and Ken Goldberg and Abhinav Gupta and Abhishek Gupta and Dinesh Jayaraman and Joseph J Lim and Jitendra Malik and Roberto Martín-Martín and Subramanian Ramamoorthy and Dorsa Sadigh and Shuran Song and Jiajun Wu and Michael C. Yip and Yuke Zhu and Thomas Kollar and Sergey Levine and Chelsea Finn},
      year={2024},
      journal={RSS},
      url_pdf = {/publication/khazatsky-2024-droid/khazatsky-2024-droid.pdf},
      url_project = {https://droid-dataset.github.io},
      url_code = {https://github.com/droid-dataset/droid},
      url_dataset = {https://huggingface.co/KarlP/droid},
      url = {https://arxiv.org/abs/2403.12945},
}
@article{stern2024physical,
  title={Training self-learning circuits for power-efficient solutions},
  author={Stern, Menachem and Dillavou, Sam and Jayaraman, Dinesh and Durian, Douglas J and Liu, Andrea J},
  journal={Applied Physics Letters (APL) Machine Learning},
  year={2024},
  abstract = {As the size and ubiquity of artificial intelligence and computational machine learning (ML) models grow, the energy required to train and use them is rapidly becoming economically and environmentally unsustainable. Recent laboratory prototypes of self-learning electronic circuits, examples of ``physical learning machines," open the door to analog hardware that directly employs physics to learn desired functions from examples at low energy cost. In this work, we show that this hardware platform allows for even further reduction of energy consumption by using good initial conditions as well as a new learning algorithm. Using analytical calculations, simulation and experiment, we show that a trade-off emerges when learning dynamics attempt to minimize both the error and the power consumption of the solution--greater power reductions can be achieved at the cost of decreasing solution accuracy. Finally, we demonstrate a practical procedure to weigh the relative importance of error and power minimization, improving power efficiency given a specific tolerance to error.},
  url_pdf = {/publication/stern-2024-physical/stern-2024-physical.pdf},
}
@article{zhang2024universal,
      title={Universal Visual Decomposer: Long-Horizon Manipulation Made Easy}, 
      author={Zichen Zhang and Yunshuang Li and Osbert Bastani and Abhishek Gupta and Dinesh Jayaraman and Yecheng Jason Ma and Luca Weihs},
      year={2024},
      journal={ICRA},
      abstract = {Real-world robotic tasks stretch over extended horizons and encompass multiple stages. Learning long-horizon manipulation tasks, however, is a long-standing challenge, and demands decomposing the overarching task into several manageable subtasks to facilitate policy learning and generalization to unseen tasks. Prior task decomposition methods require task-specific knowledge, are computationally intensive, and cannot readily be applied to new tasks. To address these shortcomings, we propose Universal Visual Decomposer (UVD), an off-the-shelf task decomposition method for visual long horizon manipulation using pre-trained visual representations designed for robotic control. At a high level, UVD discovers subgoals by detecting phase shifts in the embedding space of the pre-trained representation. Operating purely on visual demonstrations without auxiliary information, UVD can effectively extract visual subgoals embedded in the videos, while incurring zero additional training cost on top of standard visuomotor policy training. Goal-conditioned policies learned with UVD-discovered subgoals exhibit significantly improved compositional generalization at test time to unseen tasks. Furthermore, UVD-discovered subgoals can be used to construct goal-based reward shaping that jump-starts temporally extended exploration for reinforcement learning. We extensively evaluate UVD on both simulation and real-world tasks, and in all cases, UVD substantially outperforms baselines across imitation and reinforcement learning settings on in-domain and out-of-domain task sequences alike, validating the clear advantage of automated visual task decomposition within the simple, compact UVD framework.},
      url_pdf = {/publication/zhang-2024-universal/zhang-2024-universal.pdf},
      url_project = {https://zcczhang.github.io/UVD/},
      url_code = {https://github.com/zcczhang/UVD/},
      url = {https://arxiv.org/abs/2310.08581},
}
@article{qian2024soft,
  title={Recasting Generic Pretrained Vision Transformers As Object-Centric Scene Encoders For Manipulation Policies },
    author={Jianing Qian and Anastasios Panagopoulos and Dinesh Jayaraman},
  abstract={Generic re-usable pre-trained image representation encoders
have become a standard component of methods for many
computer vision tasks. As visual representations for robots
however, their utility has been limited, leading to a
recent wave of efforts to pre-train robotics-specific image
encoders that are better suited to robotic tasks than their
generic counterparts. We propose SOFT, a wrapper around
pre-trained vision transformer PVT models that bridges this
gap without any further training. Rather than construct
representations out of only the final layer activations,
SOFT individuates and locates object-like entities from PVT
attentions, and describes them with PVT activations,
producing an object-centric representation. Across standard
choices of generic pre-trained vision transformers PVT, we
demonstrate in each case that policies trained on SOFT(PVT)
far outstrip standard PVT representations for manipulation
tasks in simulated and real settings, approaching the
state-of-the-art robotics-aware representations.},
  url={https://sites.google.com/view/robot-soft/},
  journal={ICRA},
  year={2024},
  url_pdf = {/publication/qian-2024-soft/qian-2024-soft.pdf},
  url_project = {https://sites.google.com/view/robot-soft/},
  url_arxiv = {https://arxiv.org/abs/2405.15916},
}
@article{shi2024plug,
  title={Composing Pre-Trained Object-Centric Representations for Robotics From “What” and “Where” Foundation Models},
  author={Shi*, Junyao and Qian*, Jianing and Ma, Yecheng Jason and Jayaraman, Dinesh},
  abstract={There have recently been large advances both in pre-training visual representations for robotic control and segmenting unknown category objects in general images. To leverage these for improved robot learning, we propose POCR, a new framework for building pre-trained object-centric representations for robotic control. Building on theories of “what-where” representations in psychology and computer vision, we use segmentations from a pre-trained model to stably locate across timesteps, various entities in the scene, capturing “where” information. To each such segmented entity, we apply other pre-trained models that build vector descriptions suitable for robotic control tasks, thus capturing “what” the entity is. Thus, our pre-trained object-centric representations for control are constructed by appropriately combining the outputs of off-the-shelf pre-trained models, with no new training. On various simulated and real robotic tasks, we show that imitation policies for robotic manipulators trained on POCR achieve better performance and systematic generalization than state of the art pre-trained representations for robotics, as well as prior object-centric representations that are typically trained from scratch.},
url={https://sites.google.com/view/pocr},
  journal={ICRA},
  year={2024},
  url_pdf = {/publication/shi-2024-plug/shi-2024-plug.pdf},
  url_project = {https://sites.google.com/view/pocr},
  url_arxiv = {https://arxiv.org/abs/2404.13474},
}
@article{narayanan2024long,
  title={Long-HOT: A Modular Hierarchical Approach for Long-Horizon Object Transport},
  author={Narayanan, Sriram and Jayaraman, Dinesh and Chandraker, Manmohan},
  journal={ICRA},
  abstract={We aim to address key challenges in long-horizon
embodied exploration and navigation by proposing a
long-horizon object transport task called Long-HOT and a
novel modular framework for temporally extended navigation.
Agents in Long-HOT need to efficiently find and pick up
target objects that are scattered in the environment, carry
them to a goal location with load constraints, and
optionally have access to a container. We propose a modular
topological graph-based transport policy (HTP) that
explores efficiently with the help of weighted frontiers.
Our approach uses a combination of motion planning to reach
point goals within explored locations and object navigation
policies for moving towards semantic targets at unknown
locations. Experiments on both our proposed Habitat
transport task and on MultiOn benchmarks show that our
method outperforms baselines and prior works. Further, we
analyze the agent's behavior for the usage of the container
and demonstrate meaningful generalization to much harder
transport scenes with training only on simpler versions of
the task. We will release all the code and data.},
  year={2024},
  url_pdf = {/publication/narayanan-2024-long/narayanan-2024-long.pdf},
  url = {https://arxiv.org/abs/2210.15908},
}
@article{hu2024scaffolder,
  title={Privileged Sensing Scaffolds Reinforcement Learning},
  author={Edward Hu and James Springer and Oleh Rybkin and Dinesh Jayaraman},
  journal={ICLR},
  year={2024},
  abstract={We need to look at our shoelaces as we first learn to tie them but having mastered this skill, can do it from touch alone. We call this phenomenon “sensory scaffolding”: observation streams that are not needed by a master might yet aid a novice learner. We consider such sensory scaffolding setups for training artificial agents. For example, a robot arm may need to be deployed with just a low-cost, robust, general-purpose camera; yet its performance may improve by having privileged training-time-only access to informative albeit expensive and unwieldy motion capture rigs or fragile tactile sensors. For these settings, we propose Scaffolder, a reinforcement learning approach which effectively exploits privileged sensing in critics, world models, reward estimators, and other such auxiliary components that are only used at training time, to improve the target policy. For evaluating sensory scaffolding agents, we design a new “S3” suite of ten diverse simulated robotic tasks that explore a wide range of practical sensor setups. Agents must use privileged camera sensing to train blind hurdlers, privileged active visual perception to help robot arms overcome visual occlusions, privileged touch sensors to train robot hands, and more. Scaffolder easily outperforms relevant prior baselines and frequently performs comparably even to policies that have test-time access to the privileged sensors.},
  url_pdf = {/publication/hu-2024-scaffolder/hu-2024-scaffolder.pdf},
  url_project = {https://penn-pal-lab.github.io/scaffolder/},
  url_code = {https://github.com/penn-pal-lab/scaffolder},
  url_openreview = {https://openreview.net/forum?id=EpVe8jAjdx},
}
@article{wen2024relativit,
  title={Can Transformers Capture Spatial Relations between Objects?},
  author={Chuan Wen and Dinesh Jayaraman and Yang Gao},
  journal={ICLR},
  year={2024},
  abstract={Spatial relationships between objects represent key scene information for humans to understand and interact with the world. To study the capability of current computer vision systems to recognize physically grounded spatial relations, we start by proposing precise relation definitions that permit consistently annotating a benchmark dataset. Despite the apparent simplicity of this task relative to others in the recognition literature, we observe that existing approaches perform poorly on this benchmark. We propose new approaches exploiting the long-range attention capabilities of transformers for this task, and evaluating key design principles. We identify a simple "RelatiViT" architecture and demonstrate that it outperforms all current approaches. To our knowledge, this is the first method to convincingly outperform naive baselines on spatial relation prediction in in-the-wild settings.},
  url_pdf = {/publication/wen-2024-relativit/wen-2024-relativit.pdf},
  url_project = {https://sites.google.com/view/spatial-relation/},
  url_code = {https://github.com/AlvinWen428/spatial-relation-benchmark},
  url_openreview = {https://openreview.net/forum?id=HgZUcwFhjr},
  url = {https://arxiv.org/abs/2403.00729},
}
@article{ma2024eureka,
      title={Eureka: Human-Level Reward Design via Coding Large Language Models}, 
      author={Yecheng Jason Ma and William Liang and Guanzhi Wang and De-An Huang and Osbert Bastani and Dinesh Jayaraman and Yuke Zhu and Linxi Fan and Anima Anandkumar},
      year={2024},
      journal={ICLR},
      abstract = {Large Language Models (LLMs) have excelled as high-level semantic planners for sequential decision-making tasks. However, harnessing them to learn complex low-level manipulation tasks, such as dexterous pen spinning, remains an open problem. We bridge this fundamental gap and present Eureka, a human-level reward design algorithm powered by LLMs. Eureka exploits the remarkable zero-shot generation, code-writing, and in-context improvement capabilities of state-of-the-art LLMs, such as GPT-4, to perform evolutionary optimization over reward code. The resulting rewards can then be used to acquire complex skills via reinforcement learning. Without any task-specific prompting or pre-defined reward templates, Eureka generates reward functions that outperform expert human-engineered rewards. In a diverse suite of 29 open-source RL environments that include 10 distinct robot morphologies, Eureka outperforms human experts on 83% of the tasks, leading to an average normalized improvement of 52%. The generality of Eureka also enables a new gradient-free in-context learning approach to reinforcement learning from human feedback (RLHF), readily incorporating human inputs to improve the quality and the safety of the generated rewards without model updating. Finally, using Eureka rewards in a curriculum learning setting, we demonstrate for the first time, a simulated Shadow Hand capable of performing pen spinning tricks, adeptly manipulating a pen in circles at rapid speed.},
      url_pdf = {/publication/ma-2024-eureka/ma-2024-eureka.pdf},
      url_project = {https://eureka-research.github.io/},
      url_code = {https://github.com/eureka-research/Eureka},
      url = {https://arxiv.org/abs/2310.12931},
}
@article{sridhar2024memoryconsistent,
      title={Memory-Consistent Neural Networks for Imitation Learning}, 
      author={Kaustubh Sridhar and Souradeep Dutta and Dinesh Jayaraman and James Weimer and Insup Lee},
      year={2024},
      journal={ICLR},
      abstract = {Imitation learning considerably simplifies policy synthesis compared to alternative approaches by exploiting access to expert demonstrations. For such imitation policies, errors away from the training samples are particularly critical. Even rare slip-ups in the policy action outputs can compound quickly over time, since they lead to unfamiliar future states where the policy is still more likely to err, eventually causing task failures. We revisit simple supervised “behavior cloning” for conveniently training the policy from nothing more than pre-recorded demonstrations, but carefully design the model class to counter the compounding error phenomenon. Our “memory-consistent neural network” (MCNN) outputs are hard-constrained to stay within clearly specified permissible regions anchored to prototypical “memory” training samples. We provide a guaranteed upper bound for the sub-optimality gap induced by MCNN policies. Using MCNNs on 9 imitation learning tasks, with MLP, Transformer, and Diffusion backbones, spanning dexterous robotic manipulation and driving, proprioceptive inputs and visual inputs, and varying sizes and types of demonstration data, we find large and consistent gains in performance, validating that MCNNs are better-suited than vanilla deep neural networks for imitation learning applications.},
      url_pdf = {/publication/sridhar-2024-memoryconsistent/sridhar-2024-memoryconsistent.pdf},
      url_project = {https://sites.google.com/view/mcnn-imitation},
      url_code = {https://github.com/kaustubhsridhar/MCNN},
      url = {https://arxiv.org/abs/2310.06171},
}
@article{vedder2024zeroflow,
  title={ZeroFlow: Fast Zero Label Scene Flow via Distillation},
  author={Vedder, Kyle and Peri, Neehar and Chodosh, Nathaniel and Khatri, Ishan and Eaton, Eric and Jayaraman, Dinesh and Liu, Yang and Ramanan, Deva and Hays, James},
  journal={ICLR},
  year={2024},
  url_pdf = {/publication/vedder-2024-zeroflow/vedder-2024-zeroflow.pdf},
  url_project = {https://vedder.io/zeroflow.html},
  url_code = {https://github.com/kylevedder/zeroflow},
  url = {https://arxiv.org/abs/2305.10424},
}
@article{wan2024tlcontrol,
      title={TLControl: Trajectory and Language Control for Human Motion Synthesis}, 
      author={Weilin Wan and Zhiyang Dou and Taku Komura and Wenping Wang and Dinesh Jayaraman and Lingjie Liu},
      year={2024},
      abstract={Controllable human motion synthesis is essential for applications in AR/VR, gaming and embodied AI. Existing methods often focus solely on either language or full trajectory control, lacking precision in synthesizing motions aligned with user-specified trajectories, especially for multi-joint control. To address these issues, we present TLControl, a novel method for realistic human motion synthesis, incorporating both low-level Trajectory and high-level Language semantics controls, through the integration of neural-based and optimization-based techniques. Specifically, we begin with training a VQ-VAE for a compact and well-structured latent motion space organized by body parts. We then propose a Masked Trajectories Transformer (MTT) for predicting a motion distribution conditioned on language and trajectory. Once trained, we use MTT to sample initial motion predictions given user-specified partial trajectories and text descriptions as conditioning. Finally, we introduce a test-time optimization to refine these coarse predictions for precise trajectory control, which offers flexibility by allowing users to specify various optimization goals and ensures high runtime efficiency. Comprehensive experiments show that TLControl significantly outperforms the state-of-the-art in trajectory accuracy and time efficiency, making it practical for interactive and high-quality animation generation.},
      journal={ECCV},
      url_pdf = {/publication/wan-2024-tlcontrol/wan-2024-tlcontrol.pdf},
      url_project = {https://tlcontrol.weilinwl.com/},
      url_code = {https://github.com/HiWilliamWWL/TLControl},
      url_video = {https://www.youtube.com/watch?v=l9HjddeaJps},
      url = {https://arxiv.org/abs/2311.17135},
}

@InProceedings{desilva23prospective,
  title = 	 {Prospective Learning: Principled Extrapolation to the Future},
  author =       {De Silva, Ashwin and Ramesh, Rahul and Ungar, Lyle and Shuler, Marshall Hussain and Cowan, Noah J. and Platt, Michael and Li, Chen and Isik, Leyla and Roh, Seung-Eon and Charles, Adam and Venkataraman, Archana and Caffo, Brian and How, Javier J. and Kebschull, Justus M and Krakauer, John W. and Bichuch, Maxim and Kinfu, Kaleab Alemayehu and Yezerets, Eva and Jayaraman, Dinesh and Shin, Jong M. and Villar, Soledad and Phillips, Ian and Priebe, Carey E. and Hartung, Thomas and Miller, Michael I. and Dey, Jayanta and Huang, Ningyuan and Eaton, Eric and Etienne-Cummings, Ralph and Ogburn, Elizabeth L. and Burns, Randal and Osuagwu, Onyema and Mensh, Brett and Muotri, Alysson R. and Brown, Julia and White, Chris and Yang, Weiwei and Verstynen, Andrei A. Rusu Timothy and Kording, Konrad P. and Chaudhari, Pratik and Vogelstein, Joshua T.},
  booktitle = 	 {Proceedings of The 2nd Conference on Lifelong Learning Agents},
  year = 	 {2023},
  publisher =    {PMLR},
  pdf = 	 {https://proceedings.mlr.press/v232/de-silva23a/de-silva23a.pdf},
  url = 	 {https://proceedings.mlr.press/v232/de-silva23a.html},
  abstract = 	 {Learning is a process which can update decision rules, based on past experience, such that future performance improves. Traditionally, machine learning is often evaluated under the assumption that the future will be identical to the past in distribution or change adversarially. But these assumptions can be either too optimistic or pessimistic for many problems in the real world. Real world scenarios evolve over multiple spatiotemporal scales with partially predictable dynamics. Here we reformulate the learning problem to one that centers around this idea of dynamic futures that are partially learnable. We conjecture that certain sequences of tasks are not retrospectively learnable (in which the data distribution is fixed), but are prospectively learnable (in which distributions may be dynamic), suggesting that prospective learning is more difficult in kind than retrospective learning. We argue that prospective learning more accurately characterizes many real world problems that (1) currently stymie existing artificial intelligence solutions and/or (2) lack adequate explanations for how natural intelligences solve them. Thus, studying prospective learning will lead to deeper insights and solutions to currently vexing challenges in both natural and artificial intelligences.},
  url_pdf = {/publication/desilva-23-prospective/desilva-23-prospective.pdf},
  url_video = {https://youtu.be/DOgBhzs7YKs},
}
@article{kim2023im2contact,
  title={Vision-Based Contact Localization Without Touch or Force Sensing},
  author={Leon Kim and Yunshuang Li and Michael Posa and Dinesh Jayaraman},
  journal={CORL},
  year={2023},
  abstract = {Contacts play a critical role in most manipulation tasks. Robots today mainly use proximal touch/force sensors to sense contacts, but the information they provide must be calibrated and is inherently local, with practical applications relying either on extensive surface coverage or restrictive assumptions to resolve ambiguities. We propose a vision-based extrinsic contact localization task: with only a single RGB-D camera view of a robot workspace, identify when and where an object held by the robot contacts the rest of the environment. We show that careful task-attuned design is critical for a neural network trained in simulation to discover solutions that transfer well to a real robot. Our final approach im2contact demonstrates the promise of versatile general-purpose contact perception from vision alone, performing well for localizing various contact types (point, line, or planar; sticking, sliding, or rolling; single or multiple), and even under occlusions in its camera view},
  url_pdf = {/publication/kim-2023-im-2-contact/kim-2023-im-2-contact.pdf},
  url_project = {https://sites.google.com/view/im2contact/home},
}
@article{ma2023liv,
  title         = "{LIV}: Language-Image Representations and Rewards for Robotic Control",
  author        = "Yecheng Jason Ma and Vikash Kumar and Amy Zhang and Osbert Bastani and Dinesh Jayaraman",
  journal= {ICML}, 
  year = {2023},
  abstract = {We present Language-Image Value learning (LIV), a unified objective for vision-language representation and reward learning from action-free videos with text annotations. Exploiting a novel connection between dual reinforcement learning and mutual information contrastive learning, the LIV objective trains a multi-modal representation that implicitly encodes a universal value function for tasks specified as language or image goals. We use LIV to pre-train the first control-centric vision-language representation from large human video datasets such as EpicKitchen. Given only a language or image goal, the pre-trained LIV model can assign dense rewards to each frame in videos of unseen robots or humans attempting that task in unseen environments. Further, when some target domain-specific data is available, the same objective can be used to fine-tune and improve LIV and even other pre-trained representations for robotic control and reward specification in that domain. In our experiments on several simulated and real-world robot environments, LIV models consistently outperform the best prior input state representations for imitation learning, as well as reward specification methods for policy synthesis. Our results validate the advantages of joint vision-language representation and reward learning within the unified, compact LIV framework.},
  url_pdf = {/publication/ma-2023-liv/ma-2023-liv.pdf},
  url_project = {https://penn-pal-lab.github.io/LIV/},
  url_code = {https://github.com/penn-pal-lab/LIV/},
  url_dataset = {https://github.com/penn-pal-lab/LIV/blob/main/liv/dataset/README.md},
  url = {https://arxiv.org/abs/2306.00958},
}
@article{jia2024learning,
  title={Learning a Meta-Controller for Dynamic Grasping},
  author={Jia, Yinsen and Xu, Jingxi and Jayaraman, Dinesh and Song, Shuran},
  abstract={Grasping moving objects is a challenging task that requires multiple submodules such as object pose predictor, arm motion planner, etc. Each submodule operates under its own set of meta-parameters. For example, how far the pose predictor should look into the future (i.e., \textit{look-ahead time}) and the maximum amount of time the motion planner can spend planning a motion (i.e., \textit{time budget}). Many previous works assign fixed values to these parameters; however, at different moments \textit{within} a single episode of dynamic grasping, the optimal values should vary depending on the current scene. In this work, we propose a dynamic grasping pipeline with a meta-controller that controls the look-ahead time and time budget dynamically. We learn the meta-controller through reinforcement learning with a sparse reward. Our experiments show the meta-controller improves the grasping success rate (up to 28\% in the most cluttered environment) and reduces grasping time, compared to the strongest baseline. Our meta-controller learns to reason about the reachable workspace and maintain the predicted pose within the reachable region. In addition, it assigns a small but sufficient time budget for the motion planner. Our method can handle different objects, trajectories, and obstacles. Despite being trained only with 3-6 random cuboidal obstacles, our meta-controller generalizes well to 7-9 obstacles and more realistic out-of-domain household setups with unseen obstacle shapes.},
  journal={CASE},
  year={2024},
  url_pdf = {/publication/jia-2024-learning/jia-2024-learning.pdf},
  url_project = {https://yjia.net/meta},
  url_video = {https://www.youtube.com/watch?v=LbJq8EpQROI},
  url = {https://arxiv.org/abs/2302.08463},
}
@article{hu2023peg,
  title = "Planning Goals for Exploration",
  author = "Edward Hu and Richard Chang and Oleh Rybkin and Dinesh Jayaraman",
  journal= {ICLR (top 25 percent) and Best Workshop Paper at CORL 2022 Robot Adaptation Workshop}, 
  year = 2023,
  abstract = {Dropped into an unknown environment, what should an agent do to quickly learn about the environment and how to accomplish diverse tasks within it? We address this question within the goal-conditioned reinforcement learning paradigm, by identifying how the agent should set its goals at training time to maximize exploration. We propose "Planning Exploratory Goals" (PEG), a method that sets goals for each training episode to directly optimize an intrinsic exploration reward. PEG first chooses goal commands such that the agent's goal-conditioned policy, at its current level of training, will end up in states with high exploration potential. It then launches an exploration policy starting at those promising states. To enable this direct optimization, PEG learns world models and adapts sampling-based planning algorithms to "plan goal commands". In challenging simulated robotics environments including a multi-legged ant robot in a maze, and a robot arm on a cluttered tabletop, PEG exploration enables more efficient and effective training of goal-conditioned policies relative to baselines and ablations. Our ant successfully navigates a long maze, and the robot arm successfully builds a stack of three blocks upon command.},
  url_pdf = {/publication/hu-2023-peg/hu-2023-peg.pdf},
  url_project = {https://sites.google.com/view/exploratory-goals},
  url_code = {https://github.com/penn-pal-lab/peg},
  url_video = {https://www.youtube.com/watch?v=eJjvlvsFitQ},
  url_openreview = {https://openreview.net/forum?id=6qeBuZSo7Pr},
  url = {https://arxiv.org/abs/2303.13002},
}
@article{kausik2023tom,
  title         = "Learning Policy-Aware Models for Model-Based Reinforcement Learning via Transition Occupancy Matching",
  author        = "Yecheng Jason Ma and Kausik Sivakumar and Jason Yen and Osbert Bastani and Dinesh Jayaraman",
  journal = {L4DC},
  year          =  2023,
  abstract = {Standard model-based reinforcement learning (MBRL) approaches fit a transition model of the environment to all past experience, but this wastes model capacity on data that is irrelevant for policy improvement. We instead propose a new "transition occupancy matching" (TOM) objective for MBRL model learning: a model is good to the extent that the current policy experiences the same distribution of transitions inside the model as in the real environment. We derive TOM directly from a novel lower bound on the standard reinforcement learning objective. To optimize TOM, we show how to reduce it to a form of importance weighted maximum-likelihood estimation, where the automatically computed importance weights identify policy-relevant past experiences from a replay buffer, enabling stable optimization. TOM thus offers a plug-and-play model learning sub-routine that is compatible with any backbone MBRL algorithm. On various Mujoco continuous robotic control tasks, we show that TOM successfully focuses model learning on policy-relevant experience and drives policies faster to higher task rewards than alternative model learning approaches.},
  url_pdf = {/publication/kausik-2023-tom/kausik-2023-tom.pdf},
  url_project = {https://penn-pal-lab.github.io/TOM/},
  url_code = {https://github.com/kausiksivakumar/TOM},
  url = {https://arxiv.org/abs/2305.12663},
}
@article{ma2023vip,
  title         = "{VIP}: Towards Universal Visual Reward and Representation
                   via {Value-Implicit} {Pre-Training}",
  author        = "Ma, Yecheng Jason and Sodhani, Shagun and Jayaraman, Dinesh
                   and Bastani, Osbert and Kumar, Vikash and Zhang, Amy",
  journal= {ICLR (top 25 percent)}, 
  year = 2023,
  abstract = {Reward and representation learning are two long-standing challenges for learning an expanding set of robot manipulation skills from sensory observations. Given the inherent cost and scarcity of in-domain, task-specific robot data, learning from large, diverse, offline human videos has emerged as a promising path towards acquiring a generally useful visual representation for control; however, how these human videos can be used for general-purpose reward learning remains an open question. We introduce Value-Implicit Pre-training (VIP), a self-supervised pre-trained visual representation capable of generating dense and smooth reward functions for unseen robotic tasks. VIP casts representation learning from human videos as an offline goal-conditioned reinforcement learning problem and derives a self-supervised dual goal-conditioned value-function objective that does not depend on actions, enabling pre-training on unlabeled human videos. Theoretically, VIP can be understood as a novel implicit time contrastive objective that generates a temporally smooth embedding, enabling the value function to be implicitly defined via the embedding distance, which can then be used to construct the reward for any goal-image specified downstream task. Trained on large-scale Ego4D human videos and without any fine-tuning on in-domain, task-specific data, VIP's frozen representation can provide dense visual reward for an extensive set of simulated and real-robot tasks, enabling diverse reward-based visual control methods and significantly outperforming all prior pre-trained representations. Notably, VIP can enable simple, few-shot offline RL on a suite of real-world robot tasks with as few as 20 trajectories.},
  url_pdf = {/publication/ma-2023-vip/ma-2023-vip.pdf},
  url_project = {https://sites.google.com/view/vip-rl},
  url_code = {https://github.com/facebookresearch/vip},
  url = {https://arxiv.org/abs/2210.00030},
}
@article{huang2022lirf,
      title={Training Robots to Evaluate Robots: Example-Based Interactive Reward Functions for Policy Learning}, 
      author={Kun Huang and Edward Hu and Dinesh Jayaraman},
      year={2022},
      journal= {CORL},
      abstract = {Physical interactions can often help reveal information that is not readily apparent. For example, we may tug at a table leg to evaluate whether it is built well, or turn a water bottle upside down to check that it is watertight. We propose to train robots to acquire such interactive behaviors automatically, for the purpose of evaluating the result of an attempted robotic skill execution. These evaluations in turn serve as "interactive reward functions" (IRFs) for training reinforcement learning policies to perform the target skill, such as screwing the table leg tightly. In addition, even after task policies are fully trained, IRFs can serve as verification mechanisms that improve online task execution. For any given task, our IRFs can be conveniently trained using only examples of successful outcomes, and no further specification is needed to train the task policy thereafter. In our evaluations on door locking and weighted block stacking in simulation, and screw tightening on a real robot, IRFs enable large performance improvements, even outperforming baselines with access to demonstrations or carefully engineered rewards.},
      url_pdf = {/publication/huang-2022-lirf/huang-2022-lirf.pdf},
      url_project = {https://sites.google.com/view/lirf-corl-2022/},
      url_code = {https://github.com/penn-pal-lab/interactive_reward_functions},
      url_openreview = {https://openreview.net/forum?id=sK2aWU7X9b8},
      url = {https://arxiv.org/abs/2212.08961},
}
@article{qian2022dkp,
      title={Discovering Deformable Keypoint Pyramids}, 
      author={Jianing Qian and Anastasios Panagopoulos and Dinesh Jayaraman},
      year={2022},
      journal= {ECCV},
      abstract = {The locations of objects and their associated landmark keypoints can serve as versatile and semantically meaningful image representations. In natural scenes, these keypoints are often hierarchically grouped into sets corresponding to coherently moving objects and their moveable and deformable parts. Motivated by this observation, we propose Keypoint Pyramids, an approach to exploit this property for discovering keypoints without explicit supervision. Keypoint Pyramids discovers multi-level keypoint hierarchies satisfying three desiderata: comprehensiveness of the overall keypoint representation, coarse-to-fine informativeness of individual hierarchy levels, and parent-child associations of keypoints across levels. On human pose and tabletop multi-object scenes, our experimental results show that Keypoint Pyramids jointly discovers object keypoints and their natural hierarchical groupings, with finer levels adding detail to coarser levels to more comprehensively represent the visual scene. Further, we show qualitatively and quantitatively that keypoints discovered by Keypoint Pyramids using its hierarchical prior bind more consistently, and are more predictive of manually annotated semantic keypoints, compared to prior flat keypoint discovery approaches},
      url_pdf = {/publication/qian-2022-dkp/qian-2022-dkp.pdf},
      url_code = {https://github.com/jianingq/KeypointPyramids/},
}
@article{ma2022far,
  title={How Far I'll Go: Offline Goal-Conditioned Reinforcement Learning via $ f $-Advantage Regression},
  author={Ma, Yecheng Jason and Yan, Jason and Jayaraman, Dinesh and Bastani, Osbert},
  journal={NeurIPS},
  year={2022},
  abstract = {Offline goal-conditioned reinforcement learning (GCRL) promises general-purpose skill learning in the form of reaching diverse goals from purely offline datasets. We propose Goal-conditioned f-Advantage Regression (GoFAR), a novel regression-based offline GCRL algorithm derived from a state-occupancy matching perspective; the key intuition is that the goal-reaching task can be formulated as a state-occupancy matching problem between a dynamics-abiding imitator agent and an expert agent that directly teleports to the goal. In contrast to prior approaches, GoFAR does not require any hindsight relabeling and enjoys uninterleaved optimization for its value and policy networks. These distinct features confer GoFAR with much better offline performance and stability as well as statistical performance guarantee that is unattainable for prior methods. Furthermore, we demonstrate that GoFAR's training objectives can be re-purposed to learn an agent-independent goal-conditioned planner from purely offline source-domain data, which enables zero-shot transfer to new target domains. Through extensive experiments, we validate GoFAR's effectiveness in various problem settings and tasks, significantly outperforming prior state-of-art. Notably, on a real robotic dexterous manipulation task, while no other method makes meaningful progress, GoFAR acquires complex manipulation behavior that successfully accomplishes diverse goals.'},
  url_pdf = {/publication/ma-2022-far/ma-2022-far.pdf},
  url_project = {https://jasonma2016.github.io/GoFAR/},
  url_code = {https://github.com/JasonMa2016/GoFAR},
  url = {https://arxiv.org/abs/2206.03023},
}
@article{wen2022priming,
      title={Fighting Fire with Fire: Avoiding DNN Shortcuts through Priming}, 
      author={Chuan Wen and Jianing Qian and Jierui Lin and Jiaye Teng and Dinesh Jayaraman and Yang Gao},
      year={2022},
      journal= {ICML},
      abstract = {Across applications spanning supervised classification and sequential control, deep learning has been reported to find "shortcut" solutions that fail catastrophically under minor changes in the data distribution. In this paper, we show empirically that DNNs can be coaxed to avoid poor shortcuts by providing an additional "priming" feature computed from key input features, usually a coarse output estimate. Priming relies on approximate domain knowledge of these task-relevant key input features, which is often easy to obtain in practical settings. For example, one might prioritize recent frames over past frames in a video input for visual imitation learning, or salient foreground over background pixels for image classification. On NICO image classification, MuJoCo continuous control, and CARLA autonomous driving, our priming strategy works significantly better than several popular state-of-the-art approaches for feature selection and data augmentation. We connect these empirical findings to recent theoretical results on DNN optimization, and argue theoretically that priming distracts the optimizer away from poor shortcuts by creating better, simpler shortcuts.},
      url_pdf = {/publication/wen-2022-priming/wen-2022-priming.pdf},
      url_project = {https://sites.google.com/view/icml22-fighting-fire-with-fire/},
      url_code = {https://github.com/AlvinWen428/fighting-fire-with-fire},
      url = {https://arxiv.org/abs/2206.10816},
}
@article{ma2022smodice,
      title={SMODICE: Versatile Offline Imitation Learning via State Occupancy Matching}, 
      author={Yecheng Jason Ma and Andrew Shen and Dinesh Jayaraman and Osbert Bastani},
      year={2022},
      journal= {ICML},
      abstract = {We propose State Matching Offline DIstribution Correction Estimation (SMODICE), a novel and versatile algorithm for offline imitation learning (IL) via state-occupancy matching. We show that the SMODICE objective admits a simple optimization procedure through an application of Fenchel duality and an analytic solution in tabular MDPs. Without requiring access to expert actions, SMODICE can be effectively applied to three offline IL settings: (i) imitation from observations (IfO), (ii) IfO with dynamics or morphologically mismatched expert, and (iii) example-based reinforcement learning, which we show can be formulated as a state-occupancy matching problem. We extensively evaluate SMODICE on both gridworld environments as well as on high-dimensional offline benchmarks. Our results demonstrate that SMODICE is effective for all three problem settings and significantly outperforms prior state-of-art.},
      url_pdf = {/publication/ma-2022-smodice/ma-2022-smodice.pdf},
      url_project = {https://sites.google.com/view/smodice/home},
      url_code = {https://github.com/JasonMa2016/SMODICE},
      url = {https://arxiv.org/abs/2202.02433},
}
@ARTICLE{Vogelstein2022-mn,
  title         = "Prospective Learning: Back to the Future",
  author        = "Vogelstein, Joshua T and Verstynen, Timothy and Kording,
                   Konrad P and Isik, Leyla and Krakauer, John W and
                   Etienne-Cummings, Ralph and Ogburn, Elizabeth L and Priebe,
                   Carey E and Burns, Randal and Kutten, Kwame and Knierim,
                   James J and Potash, James B and Hartung, Thomas and
                   Smirnova, Lena and Worley, Paul and Savonenko, Alena and
                   Phillips, Ian and Miller, Michael I and Vidal, Rene and
                   Sulam, Jeremias and Charles, Adam and Cowan, Noah J and
                   Bichuch, Maxim and Venkataraman, Archana and Li, Chen and
                   Thakor, Nitish and Kebschull, Justus M and Albert, Marilyn
                   and Xu, Jinchong and Shuler, Marshall Hussain and Caffo,
                   Brian and Ratnanather, Tilak and Geisa, Ali and Roh,
                   Seung-Eon and Yezerets, Eva and Madhyastha, Meghana and How,
                   Javier J and Tomita, Tyler M and Dey, Jayanta and {Ningyuan}
                   and {Huang} and Shin, Jong M and Kinfu, Kaleab Alemayehu and
                   Chaudhari, Pratik and Baker, Ben and Schapiro, Anna and
                   Jayaraman, Dinesh and Eaton, Eric and Platt, Michael and
                   Ungar, Lyle and Wehbe, Leila and Kepecs, Adam and
                   Christensen, Amy and Osuagwu, Onyema and Brunton, Bing and
                   Mensh, Brett and Muotri, Alysson R and Silva, Gabriel and
                   Puppo, Francesca and Engert, Florian and Hillman, Elizabeth
                   and Brown, Julia and White, Chris and Yang, Weiwei",
  month         =  jan,
  year          =  2022,
  archivePrefix = "arXiv",
  primaryClass  = "cs.LG",
  eprint        = "2201.07372",
  abstract = {Research on both natural intelligence (NI) and artificial intelligence (AI) generally assumes that the future resembles the past: intelligent agents or systems (what we call "intelligence") observe and act on the world, then use this experience to act on future experiences of the same kind. We call this "retrospective learning". For example, an intelligence may see a set of pictures of objects, along with their names, and learn to name them. A retrospective learning intelligence would merely be able to name more pictures of the same objects. We argue that this is not what true intelligence is about. In many real world problems, both NIs and AIs will have to learn for an uncertain future. Both must update their internal models to be useful for future tasks, such as naming fundamentally new objects and using these objects effectively in a new context or to achieve previously unencountered goals. This ability to learn for the future we call "prospective learning". We articulate four relevant factors that jointly define prospective learning. Continual learning enables intelligences to remember those aspects of the past which it believes will be most useful in the future. Prospective constraints (including biases and priors) facilitate the intelligence finding general solutions that will be applicable to future problems. Curiosity motivates taking actions that inform future decision making, including in previously unmet situations. Causal estimation enables learning the structure of relations that guide choosing actions for specific outcomes, even when the specific action-outcome contingencies have never been observed before. We argue that a paradigm shift from retrospective to prospective learning will enable the communities that study intelligence to unite and overcome existing bottlenecks to more effectively explain, augment, and engineer intelligences.},
  url_pdf = {/publication/vogelstein-2022-mn/vogelstein-2022-mn.pdf},
  url = {https://arxiv.org/abs/2201.07372},
}
@article{ma2022cap, title= {Conservative and Adaptive Penalty for Model-Based Safe Reinforcement Learning}, author= {Ma, Yecheng Jason and Shen, Andrew and Bastani, Osbert and Jayaraman, Dinesh}, journal= {AAAI}, year= {2022},
  abstract = {Reinforcement Learning (RL) agents in the real world must satisfy safety constraints in addition to maximizing a reward objective. Model-based RL algorithms hold promise for reducing unsafe real-world actions: they may synthesize policies that obey all constraints using simulated samples from a learned model. However, imperfect models can result in real-world constraint violations even for actions that are predicted to satisfy all constraints. We propose Conservative and Adaptive Penalty (CAP), a model-based safe RL framework that accounts for potential modeling errors by capturing model uncertainty and adaptively exploiting it to balance the reward and the cost objectives. First, CAP inflates predicted costs using an uncertainty-based penalty. Theoretically, we show that policies that satisfy this conservative cost constraint are guaranteed to also be feasible in the true environment. We further show that this guarantees the safety of all intermediate solutions during RL training. Further, CAP adaptively tunes this penalty during training using true cost feedback from the environment. We evaluate this conservative and adaptive penalty-based approach for model-based safe RL extensively on state and image-based environments. Our results demonstrate substantial gains in sample-efficiency while incurring fewer violations than prior safe RL algorithms.},
  url_pdf = {/publication/ma-2022-cap/ma-2022-cap.pdf},
  url_code = {https://github.com/Redrew/CAP},
  url = {https://arxiv.org/abs/2112.07701},
}
@article{hu2022rac, author = {Edward S. Hu and Kun Huang and Oleh Rybkin and Dinesh Jayaraman}, journal = {ICLR}, title = {Know Thyself: Transferable Visuomotor Control Through Robot-Awareness}, year = {2022},
  abstract = {Training visuomotor robot controllers from scratch on a new robot typically requires generating large amounts of robot-specific data. Could we leverage data previously collected on another robot to reduce or even completely remove this need for robot-specific data? We propose a "robot-aware" solution paradigm that exploits readily available robot "self-knowledge" such as proprioception, kinematics, and camera calibration to achieve this. First, we learn modular dynamics models that pair a transferable, robot-agnostic world dynamics module with a robot-specific, analytical robot dynamics module. Next, we set up visual planning costs that draw a distinction between the robot self and the world. Our experiments on tabletop manipulation tasks in simulation and on real robots demonstrate that these plug-in improvements dramatically boost the transferability of visuomotor controllers, even permitting zero-shot transfer onto new robots for the very first time.},
  url_pdf = {/publication/hu-2022-rac/hu-2022-rac.pdf},
  url_project = {https://hueds.github.io/rac/},
  url_code = {https://github.com/penn-pal-lab/robot_aware_control},
  url_openreview = {https://openreview.net/forum?id=o0ehFykKVtr},
  url = {https://arxiv.org/abs/2107.09047},
}
@article{ma2021conservative, title= {Conservative Offline Distributional Reinforcement Learning}, author= {Ma, Yecheng Jason and {Jayaraman}, {Dinesh} and Bastani, Osbert}, journal= {NeurIPS}, year= {2021},
  abstract = {Many reinforcement learning (RL) problems in practice are offline, learning purely from observational data. A key challenge is how to ensure the learned policy is safe, which requires quantifying the risk associated with different actions. In the online setting, distributional RL algorithms do so by learning the distribution over returns (i.e., cumulative rewards) instead of the expected return; beyond quantifying risk, they have also been shown to learn better representations for planning. We propose Conservative Offline Distributional Actor Critic (CODAC), an offline RL algorithm suitable for both risk-neutral and risk-averse domains. CODAC adapts distributional RL to the offline setting by penalizing the predicted quantiles of the return for out-of-distribution actions. We prove that CODAC learns a conservative return distribution -- in particular, for finite MDPs, CODAC converges to an uniform lower bound on the quantiles of the return distribution; our proof relies on a novel analysis of the distributional Bellman operator. In our experiments, on two challenging robot navigation tasks, CODAC successfully learns risk-averse policies using offline data collected purely from risk-neutral agents. Furthermore, CODAC is state-of-the-art on the D4RL MuJoCo benchmark in terms of both expected and risk-sensitive performance.},
  url_pdf = {/publication/ma-2021-conservative/ma-2021-conservative.pdf},
  url_code = {https://github.com/JasonMa2016/CODAC},
  url = {https://arxiv.org/abs/2107.06106},
}
@article{kolotouros2021embracing, title= {Embracing the Reconstruction Uncertainty in 3D Human Pose Estimation}, author= {Kolotouros, Nikos and Pavlakos, Georgios and {Jayaraman}, {Dinesh} and Daniilidis, Kostas}, journal= {ICCV}, year= {2021},
  url_pdf = {/publication/kolotouros-2021-embracing/kolotouros-2021-embracing.pdf},
  url_project = {https://www.seas.upenn.edu/~nkolot/projects/prohmr/},
  url_code = {https://github.com/nkolot/ProHMR},
  url = {https://arxiv.org/abs/2108.11944},
}
% - comments @article{ma2021uncertainty, title= {What Matters More and When: Epistemic or Aleatoric Uncertainty?}, author= {Ma, Yecheng and Moore, Juston and Pleiss, Geoff and {Jayaraman}, {Dinesh} and Gardner, Jacob}, journal= {(under review)}, year= {2021}}
% - comments @article{lee2021perimeter, title= {Vision-Based Perimeter Defense Via Multi-View Active Pose Estimation}, author= {Lee, Elijah and Loianno, Giuseppe and {Jayaraman}, {Dinesh} and Kumar, Vijay}, journal= {(under review)}, year= {2021}}
@article{qian2021flood, title= {Object Representations Guided By Optical Flow}, author= {Qian, Jianing and {Jayaraman}, {Dinesh}}, journal= {NeurIPS 4th Robot Learning Workshop: Self-Supervised and Lifelong Learning}, year= {2021},
  url_pdf = {/publication/qian-2021-flood/qian-2021-flood.pdf},
}
@article{wen2021keyframe, title= {Keyframe-focused visual imitation learning}, author= {Wen, Chuan and Lin, Jierui and Qian, Jianing and Gao, Yang and {Jayaraman}, {Dinesh}}, journal= {ICML}, year= {2021},
  abstract = {Imitation learning trains control policies by mimicking pre-recorded expert demonstrations. In partially observable settings, imitation policies must rely on observation histories, but many seemingly paradoxical results show better performance for policies that only access the most recent observation. Recent solutions ranging from causal graph learning to deep information bottlenecks have shown promising results, but failed to scale to realistic settings such as visual imitation. We propose a solution that outperforms these prior approaches by upweighting demonstration keyframes corresponding to expert action changepoints. This simple approach easily scales to complex visual imitation settings. Our experimental results demonstrate consistent performance improvements over all baselines on image-based Gym MuJoCo continuous control tasks. Finally, on the CARLA photorealistic vision-based urban driving simulator, we resolve a long-standing issue in behavioral cloning for driving by demonstrating effective imitation from observation histories.},
  url_pdf = {/publication/wen-2021-keyframe/wen-2021-keyframe.pdf},
  url_project = {https://sites.google.com/view/keyframe-focused-imitation},
  url_code = {https://github.com/AlvinWen428/keyframe-focused-imitation-learning},
  url = {https://arxiv.org/abs/2106.06452},
}
@article{xu2021limits, title={How Are Learned Perception-Based Controllers Impacted by the Limits of Robust Control?},journal={L4DC}, author={Jingxi Xu and Bruce Lee and Nikolai Matni and {Dinesh} {Jayaraman}}, year={2021},
  abstract = {The difficulty of optimal control problems has classically been characterized in terms of system properties such as minimum eigenvalues of controllability/observability gramians. We revisit these characterizations in the context of the increasing popularity of data-driven techniques like reinforcement learning (RL), and in control settings where input observations are high-dimensional images and transition dynamics are unknown. Specifically, we ask: to what extent are quantifiable control and perceptual difficulty metrics of a task predictive of the performance and sample complexity of data-driven controllers? We modulate two different types of partial observability in a cartpole "stick-balancing" problem -- (i) the height of one visible fixation point on the cartpole, which can be used to tune fundamental limits of performance achievable by any controller, and by (ii) the level of perception noise in the fixation point position inferred from depth or RGB images of the cartpole. In these settings, we empirically study two popular families of controllers: RL and system identification-based H∞ control, using visually estimated system state. Our results show that the fundamental limits of robust control have corresponding implications for the sample-efficiency and performance of learned perception-based controllers. Visit our project website https://jxu.ai/rl-vs-control-web for more information.},
  url_pdf = {/publication/xu-2021-limits/xu-2021-limits.pdf},
  url_project = {https://jxu.ai/rl-vs-control-web},
  url_code = {https://github.com/jingxixu/rl-vs-control},
  url = {https://arxiv.org/abs/2104.00827},
}
@article{ramakrishnan2021exploration, author = {Ramakrishnan, Santhosh K and {Jayaraman}, {Dinesh} and Grauman, Kristen}, journal = {IJCV}, title = {An exploration of embodied visual exploration}, year = {2021},
  abstract = {Embodied computer vision considers perception for robots in novel, unstructured environments. Of particular importance is the embodied visual exploration problem: how might a robot equipped with a camera scope out a new environment? Despite the progress thus far, many basic questions pertinent to this problem remain unanswered: (i) What does it mean for an agent to explore its environment well? (ii) Which methods work well, and under which assumptions and environmental settings? (iii) Where do current approaches fall short, and where might future work seek to improve? Seeking answers to these questions, we first present a taxonomy for existing visual exploration algorithms and create a standard framework for benchmarking them. We then perform a thorough empirical study of the four state-of-the-art paradigms using the proposed framework with two photorealistic simulated 3D environments, a state-of-the-art exploration architecture, and diverse evaluation metrics. Our experimental results offer insights and suggest new performance metrics and baselines for future work in visual exploration.},
  url_project = {http://vision.cs.utexas.edu/projects/exploring-exploration/},
  url_code = {https://github.com/facebookresearch/exploring_exploration},
  url = {https://arxiv.org/abs/2001.02192},
}
@article{berseth2021smirl, author={Glen Berseth and Daniel Geng and Coline Devin and Chelsea Finn and {Dinesh} {Jayaraman} and Sergey Levine}, title={{SMiRL}: Surprise Minimizing RL in Dynamic Environments}, year = {2021}, journal ={ICLR},
  abstract = {All living organisms struggle against the forces of nature to carve out niches where they can maintain homeostasis. We propose that such a search for order amidst chaos might offer a unifying principle for the emergence of useful behaviors in artificial agents. We formalize this idea into an unsupervised reinforcement learning method called surprise minimizing RL (SMiRL). SMiRL trains an agent with the objective of maximizing the probability of observed states under a model trained on previously seen states. The resulting agents can acquire proactive behaviors that seek out and maintain stable conditions, such as balancing and damage avoidance, that are closely tied to an environment's prevailing sources of entropy, such as wind, earthquakes, and other agents. We demonstrate that our surprise minimizing agents can successfully play Tetris, Doom, control a humanoid to avoid falls and navigate to escape enemy agents, without any task-specific reward supervision. We further show that SMiRL can be used together with a standard task reward to accelerate reward-driven learning.},
  url_pdf = {/publication/berseth-2021-smirl/berseth-2021-smirl.pdf},
  url_project = {https://sites.google.com/view/surpriseminimization/home},
  url_code = {https://github.com/Neo-X/SMiRL_Code},
  url_video = {https://www.youtube.com/watch?v=LLVz6wAmYeY},
  url = {https://arxiv.org/abs/1912.05510},
}
@inproceedings{chen2021covid, author= {Chen, Hui and Li, Zhao and Feng, Sheng and Wang, Anni and Richard-Greenblatt, Melissa and Hutson, Emily and Andrianus, Stefen and Glaser, Laurel J. and Rodino, Kyle G. and Qian, Jianing and {Jayaraman}, {Dinesh} and Collman, Ronald G. and Glascock, Abigail and Bushman, Frederic D. and Lee, Jae Seung and Cherry, Sara and Fausto, Alejandra and Weiss, Susan R. and Koo, Hyun and Corby, Patricia M. and O{\textquoteright}Doherty, Una and Garfall, Alfred L. and Vogl, Dan T. and Stadtmauer, Edward A. and Wang, Ping}, title= {Femtomolar SARS-CoV-2 Antigen Detection Using the Microbubbling Digital Assay with Smartphone Readout Enables Antigen Burden Quantitation and Dynamics Tracking}, elocation-id= {2021.03.17.21253847}, year= {2021}, doi= {10.1101/2021.03.17.21253847}, publisher= {Cold Spring Harbor Laboratory Press}, abstract= {Background Little is known about the dynamics of SARS-CoV-2 antigen burden in respiratory samples in different patient populations at different stages of infection. Current rapid antigen tests cannot quantitate and track antigen dynamics with high sensitivity and specificity in respiratory samples.Methods We developed and validated an ultra-sensitive SARS-CoV-2 antigen assay with smartphone readout using the Microbubbling Digital Assay previously developed by our group, which is a platform that enables highly sensitive detection and quantitation of protein biomarkers. A computer vision-based algorithm was developed for microbubble smartphone image recognition and quantitation. A machine learning-based classifier was developed to classify the smartphone images based on detected microbubbles. Using this assay, we tracked antigen dynamics in serial swab samples from COVID patients hospitalized in ICU and immunocompromised COVID patients.Results The limit of detection (LOD) of the Microbubbling SARS-CoV-2 Antigen Assay was 0.5 pg/mL (10.6 fM) recombinant nucleocapsid (N) antigen or 4000 copies/mL inactivated SARS-CoV-2 virus in nasopharyngeal (NP) swabs, comparable to many rRT-PCR methods. The assay had high analytical specificity towards SARS-CoV-2. Compared to EUA-approved rRT-PCR methods, the Microbubbling Antigen Assay demonstrated a positive percent agreement (PPA) of 97\% (95\% confidence interval (CI), 92-99\%) in symptomatic individuals within 7 days of symptom onset and positive SARS-CoV-2 nucleic acid results, and a negative percent agreement (NPA) of 97\% (95\% CI, 94-100\%) in symptomatic and asymptomatic individuals with negative nucleic acid results. Antigen positivity rate in NP swabs gradually decreased as days-after-symptom-onset increased, despite persistent nucleic acid positivity of the same samples. The computer vision and machine learning-based automatic microbubble image classifier could accurately identify positives and negatives, based on microbubble counts and sizes. Total microbubble volume, a potential marker of antigen burden, correlated inversely with Ct values and days-after-symptom-onset. Antigen was detected for longer periods of time in immunocompromised patients with hematologic malignancies, compared to immunocompetent individuals. Simultaneous detectable antigens and nucleic acids may indicate the presence of replicating viruses in patients with persistent infections.Conclusions The Microbubbling SARS-CoV-2 Antigen Assay enables sensitive and specific detection of acute infections, and quantitation and tracking of antigen dynamics in different patient populations at various stages of infection. With smartphone compatibility and automated image processing, the assay is well-positioned to be adapted for point-of-care diagnosis and to explore the clinical implications of antigen dynamics in future studies.Competing Interest StatementThe authors have declared no competing interest.Funding StatementHC, ZL and PW have received support from National Institute of Health grants R01DA035868, R01EB029363 and National Science Foundation grant 1928334. SRW has received support from National Institute of Health grant R01AI40442 and Penn Center for Research on Coronaviruses and Other Emerging Pathogens. We thank the RADx-Tech Program, Penn Center for Precision Medicine, Penn Health-Tech and Penn Center for Innovation \& Precision Dentistry for providing funding for this project. This work was carried out in part at the Singh Center for Nanotechnology, part of the National Nanotechnology Coordinated Infrastructure Program, which is supported by the National Science Foundation grant NNCI-2025608.Author DeclarationsI confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained.YesThe details of the IRB/oversight body that provided approval or exemption for the research described are given below:The study was approved by the Institutional Review Board of the University of Pennsylvania.All necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived.YesI understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance).YesI have followed all appropriate research reporting guidelines and uploaded the relevant EQUATOR Network research reporting checklist(s) and other pertinent material as supplementary files, if applicable.YesAll data referred to in the manuscript is available from the research team. The OpenCV-based Python code for the computer vision and machine learning pipeline is available at the following address: https://github.com/jianingq/microbuble-detection-and-classification.git.}, URL= {https://www.medrxiv.org/content/early/2021/03/26/2021.03.17.21253847}, eprint= {https://www.medrxiv.org/content/early/2021/03/26/2021.03.17.21253847.full.pdf}, journal= {medRxiv},
  abstract = {Background Little is known about the dynamics of SARS-CoV-2 antigen burden in respiratory samples in different patient populations at different stages of infection. Current rapid antigen tests cannot quantitate and track antigen dynamics with high sensitivity and specificity in respiratory samples.Methods We developed and validated an ultra-sensitive SARS-CoV-2 antigen assay with smartphone readout using the Microbubbling Digital Assay previously developed by our group, which is a platform that enables highly sensitive detection and quantitation of protein biomarkers. A computer vision-based algorithm was developed for microbubble smartphone image recognition and quantitation. A machine learning-based classifier was developed to classify the smartphone images based on detected microbubbles. Using this assay, we tracked antigen dynamics in serial swab samples from COVID patients hospitalized in ICU and immunocompromised COVID patients.Results The limit of detection (LOD) of the Microbubbling SARS-CoV-2 Antigen Assay was 0.5 pg/mL (10.6 fM) recombinant nucleocapsid (N) antigen or 4000 copies/mL inactivated SARS-CoV-2 virus in nasopharyngeal (NP) swabs, comparable to many rRT-PCR methods. The assay had high analytical specificity towards SARS-CoV-2. Compared to EUA-approved rRT-PCR methods, the Microbubbling Antigen Assay demonstrated a positive percent agreement (PPA) of 97% (95% confidence interval (CI), 92-99%) in symptomatic individuals within 7 days of symptom onset and positive SARS-CoV-2 nucleic acid results, and a negative percent agreement (NPA) of 97% (95% CI, 94-100%) in symptomatic and asymptomatic individuals with negative nucleic acid results. Antigen positivity rate in NP swabs gradually decreased as days-after-symptom-onset increased, despite persistent nucleic acid positivity of the same samples. The computer vision and machine learning-based automatic microbubble image classifier could accurately identify positives and negatives, based on microbubble counts and sizes. Total microbubble volume, a potential marker of antigen burden, correlated inversely with Ct values and days-after-symptom-onset. Antigen was detected for longer periods of time in immunocompromised patients with hematologic malignancies, compared to immunocompetent individuals. Simultaneous detectable antigens and nucleic acids may indicate the presence of replicating viruses in patients with persistent infections.Conclusions The Microbubbling SARS-CoV-2 Antigen Assay enables sensitive and specific detection of acute infections, and quantitation and tracking of antigen dynamics in different patient populations at various stages of infection. With smartphone compatibility and automated image processing, the assay is well-positioned to be adapted for point-of-care diagnosis and to explore the clinical implications of antigen dynamics in future studies.Competing Interest StatementThe authors have declared no competing interest.Funding StatementHC, ZL and PW have received support from National Institute of Health grants R01DA035868, R01EB029363 and National Science Foundation grant 1928334. SRW has received support from National Institute of Health grant R01AI40442 and Penn Center for Research on Coronaviruses and Other Emerging Pathogens. We thank the RADx-Tech Program, Penn Center for Precision Medicine, Penn Health-Tech and Penn Center for Innovation & Precision Dentistry for providing funding for this project. This work was carried out in part at the Singh Center for Nanotechnology, part of the National Nanotechnology Coordinated Infrastructure Program, which is supported by the National Science Foundation grant NNCI-2025608.Author DeclarationsI confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained.YesThe details of the IRB/oversight body that provided approval or exemption for the research described are given below:The study was approved by the Institutional Review Board of the University of Pennsylvania.All necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived.YesI understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance).YesI have followed all appropriate research reporting guidelines and uploaded the relevant EQUATOR Network research reporting checklist(s) and other pertinent material as supplementary files, if applicable.YesAll data referred to in the manuscript is available from the research team. The OpenCV-based Python code for the computer vision and machine learning pipeline is available at the following address: https://github.com/jianingq/microbuble-detection-and-classification.git.},
  url_pdf = {/publication/chen-2021-covid/chen-2021-covid.pdf},
}
@article{das2020keypointirl, author = {Neha Das and Sarah Bechtle and Todor Davchev and {Dinesh} {Jayaraman} and Akshara Rai and Franziska Meier}, journal = {CORL}, title = {Model-Based Inverse Reinforcement Learning from Visual Demonstrations}, year = {2020},
  abstract = {Scaling model-based inverse reinforcement learning (IRL) to real robotic manipulation tasks with unknown dynamics remains an open problem. The key challenges lie in learning good dynamics models, developing algorithms that scale to high-dimensional state-spaces and being able to learn from both visual and proprioceptive demonstrations. In this work, we present a gradient-based inverse reinforcement learning framework that utilizes a pre-trained visual dynamics model to learn cost functions when given only visual human demonstrations. The learned cost functions are then used to reproduce the demonstrated behavior via visual model predictive control. We evaluate our framework on hardware on two basic object manipulation tasks.},
  url_pdf = {/publication/das-2020-keypointirl/das-2020-keypointirl.pdf},
  url_project = {https://sites.google.com/view/model-based-irl-from-vision},
  url_code = {https://github.com/facebookresearch/LearningToLearn/tree/main/mbirl},
  url_video = {https://www.youtube.com/watch?v=37XKIGn443o},
  url = {http://arxiv.org/abs/2010.09034},
}
@article{zhang2020cautious, title= {Cautious adaptation for reinforcement learning in safety-critical settings}, author= {Zhang, Jesse and Cheung, Brian and Finn, Chelsea and Levine, Sergey and {Jayaraman}, {Dinesh}}, journal= {ICML}, year= {2020},
  abstract = {We study the problem of safe adaptation: given a model trained on a variety of past experiences for some task, can this model learn to perform that task in a new situation while avoiding catastrophic failure? This problem setting occurs frequently in real-world reinforcement learning scenarios such as a vehicle adapting to drive in a new city, or a robotic drone adapting a policy trained only in simulation. While learning without catastrophic failures is exceptionally difficult, prior experience can allow us to learn models that make this much easier. These models might not directly transfer to new settings, but can enable cautious adaptation that is substantially safer than na\"{i}ve adaptation as well as learning from scratch. Building on this intuition, we propose risk-averse domain adaptation (RADA). RADA works in two steps: it first trains probabilistic model-based RL agents in a population of source domains to gain experience and capture epistemic uncertainty about the environment dynamics. Then, when dropped into a new environment, it employs a pessimistic exploration policy, selecting actions that have the best worst-case performance as forecasted by the probabilistic model. We show that this simple maximin policy accelerates domain adaptation in a safety-critical driving environment with varying vehicle sizes. We compare our approach against other approaches for adapting to new environments.},
  url_pdf = {/publication/zhang-2020-cautious/zhang-2020-cautious.pdf},
  url_project = {https://sites.google.com/berkeley.edu/carl},
  url_code = {http://proceedings.mlr.press/v119/www.github.com/jesbu1/carl},
  url = {https://arxiv.org/abs/2008.06622},
}
@article{lambeta2020digit, title= {Digit: A novel design for a low-cost compact high-resolution tactile sensor with application to in-hand manipulation}, author= {Lambeta, Mike and Chou, Po-Wei and Tian, Stephen and Yang, Brian and Maloon, Benjamin and Most, Victoria Rose and Stroud, Dave and Santos, Raymond and Byagowi, Ahmad and Kammerer, Gregg and {Jayaraman}, {Dinesh} and Calandra, Roberto}, journal= {ICRA and IEEE RA-L}, year= {2020},
  abstract = {Despite decades of research, general purpose in-hand manipulation remains one of the unsolved challenges of robotics. One of the contributing factors that limit current robotic manipulation systems is the difficulty of precisely sensing contact forces -- sensing and reasoning about contact forces are crucial to accurately control interactions with the environment. As a step towards enabling better robotic manipulation, we introduce DIGIT, an inexpensive, compact, and high-resolution tactile sensor geared towards in-hand manipulation. DIGIT improves upon past vision-based tactile sensors by miniaturizing the form factor to be mountable on multi-fingered hands, and by providing several design improvements that result in an easier, more repeatable manufacturing process, and enhanced reliability. We demonstrate the capabilities of the DIGIT sensor by training deep neural network model-based controllers to manipulate glass marbles in-hand with a multi-finger robotic hand. To provide the robotic community access to reliable and low-cost tactile sensors, we open-source the DIGIT design at www.digit.ml.},
  url_pdf = {/publication/lambeta-2020-digit/lambeta-2020-digit.pdf},
  url_project = {www.digit.ml},
  url_code = {https://github.com/facebookresearch/digit-interface},
  url_video = {https://www.youtube.com/watch?v=F73kkqiHGwE},
  url = {https://arxiv.org/abs/2005.14679},
}
@article{ma2021diverse, title= {Likelihood-Based Diverse Sampling for Trajectory Forecasting}, author= {Ma, Yecheng Jason and Inala, Jeevana Priya and {Jayaraman}, {Dinesh} and Bastani, Osbert}, journal= {ICCV}, year= {2021},
  abstract = {For autonomous cars to drive safely and effectively, they must anticipate the stochastic future trajectories of other agents in the scene, such as pedestrians and other cars. Forecasting such complex multi-modal distributions requires powerful probabilistic approaches. Normalizing flows have recently emerged as an attractive tool to model such distributions. However, when generating trajectory predictions from a flow model, a key drawback is that independent samples often do not adequately capture all the modes in the underlying distribution. We propose Diversity Sampling for Flow (DSF), a method for improving the quality and the diversity of trajectory samples from a pre-trained flow model. Rather than producing individual samples, DSF produces a set of trajectories in one shot. Given a pre-trained forecasting flow model, we train DSF using gradients from the model, to optimize an objective function that rewards high likelihood for individual trajectories in the predicted set, together with high spatial separation between trajectories. DSF is easy to implement, and we show that it offers a simple plug-in improvement for several existing flow-based forecasting models, achieving state-of-art results on two challenging vehicle and pedestrian forecasting benchmarks.},
  url_pdf = {/publication/ma-2021-diverse/ma-2021-diverse.pdf},
  url_code = {https://github.com/JasonMa2016/LDS},
  url = {https://arxiv.org/abs/2011.15084},
}
@article{wen2020copycat, title= {Fighting Copycat Agents in Behavioral Cloning from Observation Histories}, author= {Wen, Chuan and Lin, Jierui and Darrell, Trevor and {Jayaraman}, {Dinesh} and Gao, Yang}, journal= {NeurIPS}, year= {2020},
  url_pdf = {/publication/wen-2020-copycat/wen-2020-copycat.pdf},
  url = {https://arxiv.org/abs/2010.14876},
}
@article{pertsch2020long, title= {Long-horizon visual planning with goal-conditioned hierarchical predictors}, author= {Pertsch, Karl and Rybkin, Oleh and Ebert, Frederik and {Jayaraman}, {Dinesh} and Finn, Chelsea and Levine, Sergey}, journal= {NeurIPS}, year= {2020},
  abstract = {The ability to predict and plan into the future is fundamental for agents acting in the world. To reach a faraway goal, we predict trajectories at multiple timescales, first devising a coarse plan towards the goal and then gradually filling in details. In contrast, current learning approaches for visual prediction and planning fail on long-horizon tasks as they generate predictions (1) without considering goal information, and (2) at the finest temporal resolution, one step at a time. In this work we propose a framework for visual prediction and planning that is able to overcome both of these limitations. First, we formulate the problem of predicting towards a goal and propose the corresponding class of latent space goal-conditioned predictors (GCPs). GCPs significantly improve planning efficiency by constraining the search space to only those trajectories that reach the goal. Further, we show how GCPs can be naturally formulated as hierarchical models that, given two observations, predict an observation between them, and by recursively subdividing each part of the trajectory generate complete sequences. This divide-and-conquer strategy is effective at long-term prediction, and enables us to design an effective hierarchical planning algorithm that optimizes trajectories in a coarse-to-fine manner. We show that by using both goal-conditioning and hierarchical prediction, GCPs enable us to solve visual planning tasks with much longer horizon than previously possible.},
  url_pdf = {/publication/pertsch-2020-long/pertsch-2020-long.pdf},
  url_project = {https://orybkin.github.io/video-gcp/},
  url_code = {https://github.com/orybkin/video-gcp},
  url_video = {https://www.youtube.com/watch?v=axXx-x86IeY},
  url = {https://arxiv.org/abs/2006.13205},
}
@article{yang2020mavric, title={{MAVRIC}: Morphology-Agnostic Visual Robotic Control}, author={Yang*, Brian and {Jayaraman}*, {Dinesh} and Berseth, Glen and Efros, Alexei and Levine, Sergey}, journal={ICRA and IEEE RA-L}, year={2020},
  abstract = {Existing approaches for visuomotor robotic control typically require characterizing the robot in advance by calibrating the camera or performing system identification. We propose MAVRIC, an approach that works with minimal prior knowledge of the robot's morphology, and requires only a camera view containing the robot and its environment and an unknown control interface. MAVRIC revolves around a mutual information-based method for self-recognition, which discovers visual ``control points'' on the robot body within a few seconds of exploratory interaction, and these control points in turn are then used for visual servoing. MAVRIC can control robots with imprecise actuation, no proprioceptive feedback, unknown morphologies including novel tools, unknown camera poses, and even unsteady handheld cameras. We demonstrate our method on visually-guided 3D point reaching, trajectory following, and robot-to-robot imitation.},
  url_pdf = {/publication/yang-2020-mavric/yang-2020-mavric.pdf},
  url_project = {https://sites.google.com/berkeley.edu/mavric/home},
  url = {https://arxiv.org/abs/1912.13360},
}
@article{de2019causal, title= {Causal Confusion in Imitation Learning}, author= {de Haan, Pim and {Jayaraman}, {Dinesh} and Levine, Sergey}, journal= {NeurIPS}, year= {2019},
  abstract = {Behavioral cloning reduces policy learning to supervised learning by training a discriminative model to predict expert actions given observations. Such discriminative models are non-causal: the training procedure is unaware of the causal structure of the interaction between the expert and the environment. We point out that ignoring causality is particularly damaging because of the distributional shift in imitation learning. In particular, it leads to a counter-intuitive "causal confusion" phenomenon: access to more information can yield worse performance. We investigate how this problem arises, and propose a solution to combat it through targeted interventions---either environment interaction or expert queries---to determine the correct causal model. We show that causal confusion occurs in several benchmark control domains as well as realistic driving settings, and validate our solution against DAgger and other baselines and ablations.},
  url_pdf = {/publication/de-2019-causal/de-2019-causal.pdf},
  url_project = {https://sites.google.com/view/causal-confusion},
  url_code = {https://github.com/pimdh/causal-confusion},
  url_dataset = {https://drive.google.com/file/d/1QnaVmDk1HhkiNEcTBj0Mh1NC0JVYDn4o/view},
  url_video = {https://www.youtube.com/watch?v=_dh2-2b1jmU},
  url = {https://arxiv.org/abs/1905.11979},
}
@article{ramakrishnan2019emergence, title= {Emergence of exploratory look-around behaviors through active observation completion}, author= {Ramakrishnan*, Santhosh K and {Jayaraman}*, {Dinesh} and Grauman, Kristen}, journal= {Science Robotics}, year= {2019},
  abstract = {Standard computer vision systems assume access to intelligently captured inputs (e.g., photos from a human photographer), yet autonomously capturing good observations is a major challenge in itself. We address the problem of learning to look around: How can an agent learn to acquire informative visual observations? We propose a reinforcement learning solution, where the agent is rewarded for reducing its uncertainty about the unobserved portions of its environment. Specifically, the agent is trained to select a short sequence of glimpses, after which it must infer the appearance of its full environment. To address the challenge of sparse rewards, we further introduce sidekick policy learning, which exploits the asymmetry in observability between training and test time. The proposed methods learned observation policies that not only performed the completion task for which they were trained but also generalized to exhibit useful "look-around" behavior for a range of active perception tasks.},
  url_pdf = {/publication/ramakrishnan-2019-emergence/ramakrishnan-2019-emergence.pdf},
  url_project = {http://vision.cs.utexas.edu/projects/visual-exploration/},
  url_code = {https://github.com/srama2512/visual-exploration},
  url_dataset = {https://utexas.box.com/shared/static/glrybela16mi1xofkqdudfzpk2gmygr2.zip},
  url = {https://arxiv.org/abs/1906.11407},
}
@article{tian2019manipulation, title= {Manipulation by feel: Touch-based control with deep predictive models}, author= {Tian, Stephen and Ebert, Frederik and {Jayaraman}, {Dinesh} and Mudigonda, Mayur and Finn, Chelsea and Calandra, Roberto and Levine, Sergey}, journal= {ICRA}, year= {2019},
  abstract = {Touch sensing is widely acknowledged to be important for dexterous robotic manipulation, but exploiting tactile sensing for continuous, non-prehensile manipulation is challenging. General purpose control techniques that are able to effectively leverage tactile sensing as well as accurate physics models of contacts and forces remain largely elusive, and it is unclear how to even specify a desired behavior in terms of tactile percepts. In this paper, we take a step towards addressing these issues by combining high-resolution tactile sensing with data-driven modeling using deep neural network dynamics models. We propose deep tactile MPC, a framework for learning to perform tactile servoing from raw tactile sensor inputs, without manual supervision. We show that this method enables a robot equipped with a GelSight-style tactile sensor to manipulate a ball, analog stick, and 20-sided die, learning from unsupervised autonomous interaction and then using the learned tactile predictive model to reposition each object to user-specified configurations, indicated by a goal tactile reading. Videos, visualizations and the code are available at <a href=https://sites.google.com/view/deeptactilempc>this https URL</a>.},
  url_pdf = {/publication/tian-2019-manipulation/tian-2019-manipulation.pdf},
  url_project = {https://sites.google.com/view/deeptactilempc},
  url = {http://arxiv.org/abs/1903.04128},
}
@article{yang2019replab, title= {REPLAB: A reproducible low-cost arm benchmark for robotic learning}, author= {Yang, Brian and {Jayaraman}, {Dinesh} and Zhang, Jesse and Levine, Sergey}, journal= {ICRA}, year={2019},
  abstract = {Standardized evaluation measures have aided in the progress of machine learning approaches in disciplines such as computer vision and machine translation. In this paper, we make the case that robotic learning would also benefit from benchmarking, and present a template for a vision-based manipulation benchmark. Our benchmark is built on “REPLAB,” a reproducible and self-contained hardware stack (robot arm, camera, and workspace) that costs about 2000 USD and occupies a cuboid of size 70x40x60 cm. Each REPLAB cell may be assembled within a few hours. Through this low-cost, compact design, REPLAB aims to drive wide participation by lowering the barrier to entry into robotics and to enable easy scaling to many robots. We envision REPLAB as a framework for reproducible research across manipulation tasks, and as a step in this direction, we define a grasping benchmark consisting of a task definition, evaluation protocol, performance measures, and a dataset of over 50,000 grasp attempts. We implement, evaluate, and analyze several previously proposed grasping approaches to establish baselines for this benchmark. Project page with assembly instructions, additional details, and videos, is at <a href=https://goo.gl/5F9dP4>this https URL</a>.},
  url_pdf = {/publication/yang-2019-replab/yang-2019-replab.pdf},
  url_project = {https://sites.google.com/view/replab/},
  url_code = {https://github.com/bhyang/replab},
  url = {http://arxiv.org/abs/1905.07447},
}
@article{jayaraman2019time, title= {Time-agnostic prediction: Predicting predictable video frames}, author= {{Jayaraman}, {Dinesh} and Ebert, Frederik and Efros, Alexei A and Levine, Sergey}, journal= {ICLR}, year= {2019},
  abstract = {Prediction is arguably one of the most basic functions of an intelligent system. In general, the problem of predicting events in the future or between two waypoints is exceedingly difficult. However, most phenomena naturally pass through relatively predictable bottlenecks---while we cannot predict the precise trajectory of a robot arm between being at rest and holding an object up, we can be certain that it must have picked the object up. To exploit this, we decouple visual prediction from a rigid notion of time. While conventional approaches predict frames at regularly spaced temporal intervals, our time-agnostic predictors (TAP) are not tied to specific times so that they may instead discover predictable "bottleneck" frames no matter when they occur. We evaluate our approach for future and intermediate frame prediction across three robotic manipulation tasks. Our predictions are not only of higher visual quality, but also correspond to coherent semantic subgoals in temporally extended tasks.},
  url_pdf = {/publication/jayaraman-2019-time/jayaraman-2019-time.pdf},
  url_project = {https://sites.google.com/view/predictingthepredictablesupp/home},
  url = {https://arxiv.org/abs/1808.07784},
}
@misc{ma2018techniques, title= {Techniques for rectification of camera arrays}, author= {Ma, Tao and Sun, Wei and Nestares, Oscar and Seshadrinathan, Kalpana and {Jayaraman}, {Dinesh}}, year= {2018}, month= {jan~23}, note= {US Patent 9,875,543},
  url_pdf = {/publication/ma-2018-techniques/ma-2018-techniques.pdf},
}
@article{jayaraman2018end, title= {End-to-end policy learning for active visual categorization}, author= {{Jayaraman}, {Dinesh} and Grauman, Kristen}, journal= {IEEE TPAMI}, year= {2018},
  abstract = {Visual recognition systems mounted on autonomous moving agents face the challenge of unconstrained data, but simultaneously have the opportunity to improve their performance by moving to acquire new views at test time. In this work, we first show how a recurrent neural network-based system may be trained to perform end-to-end learning of motion policies suited for this “active recognition” setting. Further, we hypothesize that active vision requires an agent to have the capacity to reason about the effects of its motions on its view of the world. To verify this hypothesis, we attempt to induce this capacity in our active recognition pipeline, by simultaneously learning to forecast the effects of the agent's motions on its internal representation of the environment conditional on all past views. Results across three challenging datasets confirm both that our end-to-end system successfully learns meaningful policies for active category recognition, and that “learning to look ahead” further boosts recognition performance.},
  url_pdf = {https://ieeexplore.ieee.org/document/8367872},
  url = {jayaraman-2018-end-preprint.pdf},
}
@article{jayaraman2018learning, title= {Learning to look around: Intelligently exploring unseen environments for unknown tasks}, author= {{Jayaraman}, {Dinesh} and Grauman, Kristen}, journal= {CVPR}, year= {2018},
  abstract = {It is common to implicitly assume access to intelligently captured inputs (e.g., photos from a human photographer), yet autonomously capturing good observations is itself a major challenge. We address the problem of learning to look around: if a visual agent has the ability to voluntarily acquire new views to observe its environment, how can it learn efficient exploratory behaviors to acquire informative observations? We propose a reinforcement learning solution, where the agent is rewarded for actions that reduce its uncertainty about the unobserved portions of its environment. Based on this principle, we develop a recurrent neural network-based approach to perform active completion of panoramic natural scenes and 3D object shapes. Crucially, the learned policies are not tied to any recognition task nor to the particular semantic content seen during training. As a result, 1) the learned 'look around' behavior is relevant even for new tasks in unseen environments, and 2) training data acquisition involves no manual labeling. Through tests in diverse settings, we demonstrate that our approach learns useful generic policies that transfer to new unseen tasks and environments. Completion episodes are shown at <a href=https://goo.gl/BgWX3W>this https URL</a>.},
  url_pdf = {/publication/jayaraman-2018-learning/jayaraman-2018-learning.pdf},
  url = {https://arxiv.org/abs/1709.00507},
}
@article{calandra2018more, title= {More Than a Feeling: Learning to Grasp and Regrasp using Vision and Touch}, author= {Calandra, Roberto and Owens, Andrew and {Jayaraman}, {Dinesh} and Lin, Justin and Yuan, Wenzhen and Malik, Jitendra and Adelson, Edward H and Levine, Sergey}, journal= {IROS and IEEE RA-L}, year= {2018},
  abstract = {For humans, the process of grasping an object relies heavily on rich tactile feedback. Most recent robotic grasping work, however, has been based only on visual input, and thus cannot easily benefit from feedback after initiating contact. In this paper, we investigate how a robot can learn to use tactile information to iteratively and efficiently adjust its grasp. To this end, we propose an end-to-end action-conditional model that learns regrasping policies from raw visuo-tactile data. This model – a deep, multimodal convolutional network – predicts the outcome of a candidate grasp adjustment, and then executes a grasp by iteratively selecting the most promising actions. Our approach requires neither calibration of the tactile sensors, nor any analytical modeling of contact forces, thus reducing the engineering effort required to obtain efficient grasping policies. We train our model with data from about 6,450 grasping trials on a two-finger gripper equipped with GelSight high-resolution tactile sensors on each finger. Across extensive experiments, our approach outperforms a variety of baselines at (i) estimating grasp adjustment outcomes, (ii) selecting efficient grasp adjustments for quick grasping, and (iii) reducing the amount of force applied at the fingers, while maintaining competitive performance. Finally, we study the choices made by our model and show that it has successfully acquired useful and interpretable grasping behaviors.},
  url_pdf = {/publication/calandra-2018-more/calandra-2018-more.pdf},
  url_project = {https://sites.google.com/view/more-than-a-feeling/},
  url_video = {https://www.youtube.com/watch?v=0eQCSrKdt5w},
  url = {https://arxiv.org/abs/1805.11085},
}
@article{jayaraman2018shapecodes, title= {Shapecodes: self-supervised feature learning by lifting views to viewgrids}, author= {{Jayaraman}, {Dinesh} and Gao, Ruohan and Grauman, Kristen}, journal= {ECCV}, year= {2018},
  abstract = {We introduce an unsupervised feature learning approach that embeds 3D shape information into a single-view image representation. The main idea is a self-supervised training objective that, given only a single 2D image, requires all unseen views of the object to be predictable from learned features. We implement this idea as an encoder-decoder convolutional neural network. The network maps an input image of an unknown category and unknown viewpoint to a latent space, from which a deconvolutional decoder can best "lift" the image to its complete viewgrid showing the object from all viewing angles. Our class-agnostic training procedure encourages the representation to capture fundamental shape primitives and semantic regularities in a data-driven manner---without manual semantic labels. Our results on two widely-used shape datasets show 1) our approach successfully learns to perform "mental rotation" even for objects unseen during training, and 2) the learned latent space is a powerful representation for object recognition, outperforming several existing unsupervised feature learning methods.},
  url_pdf = {/publication/jayaraman-2018-shapecodes/jayaraman-2018-shapecodes.pdf},
  url = {https://arxiv.org/abs/1709.00505},
}
@misc{nestares2017techniques, title= {Techniques for improved focusing of camera arrays}, author= {Nestares, Oscar and Seshadrinathan, Kalpana and {Jayaraman}, {Dinesh}}, year= {2017}, month= {aug~22}, note= {US Patent 9,743,016},
  url_pdf = {/publication/nestares-2017-techniques/nestares-2017-techniques.pdf},
}
@incollection{chen2017divide, title= {Divide, share, and conquer: Multi-task attribute learning with selective sharing}, author= {{Jayaraman}, {{Dinesh}} and Chen, Chao-Yeh and Sha, Fei and Grauman, Kristen}, booktitle= {Visual attributes}, pages= {49--85}, year= {2017}, publisher= {Springer, Cham},
  abstract = {Existing methods to learn visual attributes are plagued by two common issues: (i) they are prone to confusion by properties that are correlated with the attribute of interest among training samples, and (ii) they often learn generic, imprecise "lowest common denominator" attribute models in an attempt to generalize across classes where a single attribute may have very different visual manifestations. Yet, many proposed applications of attributes rely on being able to learn the precise and correct semantic concept corresponding to each attribute. We argue that these issues are both largely due to indiscriminate "oversharing" amongst attribute classifiers along two axes — (i) visual features and (ii) classifier parameters. To address both these issues, we introduce the general idea of selective sharing during multitask learning of attributes. First, we show how selective sharing helps learn decorrelated models for each attribute in a vocabulary. Second, we show how selective sharing permits a new form of transfer learning between attributes, yielding a specialized attribute model for each individual object category. We validate both these instantiations of our selective sharing idea through extensive experiments on multiple datasets. We show how they help preserve semantics in learned attribute models, benefitting various downstream applications such as image retrieval or zero-shot learning.},
  url_pdf = {/publication/chen-2017-divide/chen-2017-divide.pdf},
}
@phdthesis{jayaraman2017embodied, title= {Embodied learning for visual recognition}, author= {{Jayaraman}, {Dinesh}}, year= {2017}, school= {UT Austin},
  url_pdf = {/publication/jayaraman-2017-embodied/jayaraman-2017-embodied.pdf},
}
@article{jayaraman2017learning, title= {Learning Image Representations Tied to Egomotion from Unlabeled Video}, author= {{Jayaraman}, {Dinesh} and Grauman, Kristen}, journal= {IJCV Special Issue of Best Papers from ICCV 2015}, year= {2017},
  abstract = {Understanding how images of objects and scenes behave in response to specific egomotions is a crucial aspect of proper visual development, yet existing visual learning methods are conspicuously disconnected from the physical source of their images. We propose a new “embodied” visual learning paradigm, exploiting proprioceptive motor signals to train visual representations from egocentric video with no manual supervision. Specifically, we enforce that our learned features exhibit equivariance i.e., they respond predictably to transformations associated with distinct egomotions. With three datasets, we show that our unsupervised feature learning approach significantly outperforms previous approaches on visual recognition and next-best-view prediction tasks. In the most challenging test, we show that features learned from video captured on an autonomous driving platform improve large-scale scene recognition in static images from a disjoint domain.},
  url_pdf = {http://rdcu.be/pOGq},
  url_project = {http://vision.cs.utexas.edu/projects/egoequiv/},
}
@article{jayaraman2016look, title= {Look-ahead before you leap: end-to-end active recognition by forecasting the effect of motion}, author= {{Jayaraman}, {Dinesh} and Grauman, Kristen}, journal= {ECCV}, year= {2016},
  abstract = {Visual recognition systems mounted on autonomous moving agents face the challenge of unconstrained data, but simultaneously have the opportunity to improve their performance by moving to acquire new views of test data. In this work, we first show how a recurrent neural network-based system may be trained to perform end-to-end learning of motion policies suited for this "active recognition" setting. Further, we hypothesize that active vision requires an agent to have the capacity to reason about the effects of its motions on its view of the world. To verify this hypothesis, we attempt to induce this capacity in our active recognition pipeline, by simultaneously learning to forecast the effects of the agent's motions on its internal representation of the environment conditional on all past views. Results across two challenging datasets confirm both that our end-to-end system successfully learns meaningful policies for active category recognition, and that "learning to look ahead" further boosts recognition performance.},
  url_pdf = {/publication/jayaraman-2016-look/jayaraman-2016-look.pdf},
  url_project = {http://vision.cs.utexas.edu/projects/lookahead_active/},
  url_video = {http://videolectures.net/eccv2016_jayaraman_active_recognition/},
  url = {http://arxiv.org/abs/1605.00164},
}
@article{gao2016object, title= {Object-Centric Representation Learning from Unlabeled Videos}, author= {Gao, Ruohan and {Jayaraman}, {Dinesh} and Grauman, Kristen}, journal= {ACCV}, year= {2016},
  abstract = {Supervised (pre-)training currently yields state-of-the-art performance for representation learning for visual recognition, yet it comes at the cost of (1) intensive manual annotations and (2) an inherent restriction in the scope of data relevant for learning. In this work, we explore unsupervised feature learning from unlabeled video. We introduce a novel object-centric approach to temporal coherence that encourages similar representations to be learned for object-like regions segmented from nearby frames. Our framework relies on a Siamese-triplet network to train a deep convolutional neural network (CNN) representation. Compared to existing temporal coherence methods, our idea has the advantage of lightweight preprocessing of the unlabeled video (no tracking required) while still being able to extract object-level regions from which to learn invariances. Furthermore, as we show in results on several standard datasets, our method typically achieves substantial accuracy gains over competing unsupervised methods for image classification and retrieval tasks.},
  url_pdf = {/publication/gao-2016-object/gao-2016-object.pdf},
  url_project = {http://vision.cs.utexas.edu/projects/object_centric_unsup/},
  url = {https://arxiv.org/abs/1612.00500},
}
@article{su2016pano2vid, title= {Pano2Vid: Automatic cinematography for watching 360-degree videos}, author= {Su, Yu-Chuan and {Jayaraman}, {Dinesh} and Grauman, Kristen}, journal= {ACCV}, year= {2016},
  abstract = {We introduce the novel task of Pano2Vid − automatic cinematography in panoramic 360∘ videos. Given a 360-degree video, the goal is to direct an imaginary camera to virtually capture natural-looking normal field-of-view (NFOV) video. By selecting "where to look" within the panorama at each time step, Pano2Vid aims to free both the videographer and the end viewer from the task of determining what to watch. Towards this goal, we first compile a dataset of 360-degree videos downloaded from the web, together with human-edited NFOV camera trajectories to facilitate evaluation. Next, we propose AutoCam, a data-driven approach to solve the Pano2Vid task. AutoCam leverages NFOV web video to discriminatively identify space-time "glimpses" of interest at each time instant, and then uses dynamic programming to select optimal human-like camera trajectories. Through experimental evaluation on multiple newly defined Pano2Vid performance measures against several baselines, we show that our method successfully produces informative videos that could conceivably have been captured by human videographers.},
  url_pdf = {/publication/su-2016-pano-2-vid/su-2016-pano-2-vid.pdf},
  url_project = {http://www.cs.utexas.edu/~ycsu/projects/Pano2Vid},
  url = {https://arxiv.org/abs/1612.02335},
}
@article{jayaraman2016slow, title= {Slow and steady feature analysis: higher order temporal coherence in video}, author= {{Jayaraman}, {Dinesh} and Grauman, Kristen}, journal= {CVPR}, year= {2016},
  abstract = {How can unlabeled video augment visual learning? Existing methods perform "slow" feature analysis, encouraging the representations of temporally close frames to exhibit only small differences. While this standard approach captures the fact that high-level visual signals change slowly over time, it fails to capture *how* the visual content changes. We propose to generalize slow feature analysis to "steady" feature analysis. The key idea is to impose a prior that higher order derivatives in the learned feature space must be small. To this end, we train a convolutional neural network with a regularizer on tuples of sequential frames from unlabeled video. It encourages feature changes over time to be smooth, i.e., similar to the most recent changes. Using five diverse datasets, including unlabeled YouTube and KITTI videos, we demonstrate our method's impact on object, scene, and action recognition tasks. We further show that our features learned from unlabeled video can even surpass a standard heavily supervised pretraining approach.},
  url_pdf = {/publication/jayaraman-2016-slow/jayaraman-2016-slow.pdf},
  url_video = {https://www.youtube.com/watch?v=1oYXbVyffgA},
  url = {http://arxiv.org/abs/1506.04714},
}
@article{jayaraman2015learning, title= {Learning image representations tied to ego-motion}, author= {{Jayaraman}, {Dinesh} and Grauman, Kristen}, journal= {ICCV}, year= {2015},
  abstract = {Understanding how images of objects and scenes behave in response to specific ego-motions is a crucial aspect of proper visual development, yet existing visual learning methods are conspicuously disconnected from the physical source of their images. We propose to exploit proprioceptive motor signals to provide unsupervised regularization in convolutional neural networks to learn visual representations from egocentric video. Specifically, we enforce that our learned features exhibit equivariance i.e. they respond predictably to transformations associated with distinct ego-motions. With three datasets, we show that our unsupervised feature learning approach significantly outperforms previous approaches on visual recognition and next-best-view prediction tasks. In the most challenging test, we show that features learned from video captured on an autonomous driving platform improve large-scale scene recognition in static images from a disjoint domain.},
  url_pdf = {/publication/jayaraman-2015-learning/jayaraman-2015-learning.pdf},
  url_project = {http://vision.cs.utexas.edu/projects/egoequiv/},
  url_video = {http://videolectures.net/iccv2015_jayaraman_image_representations/},
  url = {http://arxiv.org/abs/1505.02206},
}
@article{jayaraman2014decorrelating, title= {Decorrelating semantic visual attributes by resisting the urge to share}, author= {{Jayaraman}, {Dinesh} and Sha, Fei and Grauman, Kristen}, journal={CVPR}, year= {2014},
  abstract = {Existing methods to learn visual attributes are prone to learning the wrong thing—namely, properties that are correlated with the attribute of interest among training samples. Yet, many proposed applications of attributes rely on being able to learn the correct semantic concept corresponding to each attribute. We propose to resolve such confusions by jointly learning decorrelated, discriminative attribute models. Leveraging side information about semantic relatedness, we develop a multi-task learning approach that uses structured sparsity to encourage feature competition among unrelated attributes and feature sharing among related attributes. On three challenging datasets, we show that accounting for structure in the visual attribute space is key to learning attribute models that preserve semantics, yielding improved generalizability that helps in the recognition and discovery of unseen object categories.},
  url_pdf = {/publication/jayaraman-2014-decorrelating/jayaraman-2014-decorrelating.pdf},
  url_project = {http://vision.cs.utexas.edu/projects/resistshare/},
  url_video = {http://techtalks.tv/talks/decorrelating-semantic-visual-attributes-by-resisting-the-urge-to-share/60279/},
}
@article{jayaraman2014zero, title= {Zero-shot recognition with unreliable attributes}, author= {{Jayaraman}, {Dinesh} and Grauman, Kristen}, journal= {NeurIPS}, year= {2014},
  abstract = {In principle, zero-shot learning makes it possible to train a recognition model simply by specifying the category's attributes. For example, with classifiers for generic attributes like "striped" and "four-legged", one can construct a classifier for the zebra category by enumerating which properties it possesses---even without providing zebra training images. In practice, however, the standard zero-shot paradigm suffers because attribute predictions in novel images are hard to get right. We propose a novel random forest approach to train zero-shot models that explicitly accounts for the unreliability of attribute predictions. By leveraging statistics about each attribute's error tendencies, our method obtains more robust discriminative models for the unseen classes. We further devise extensions to handle the few-shot scenario and unreliable attribute descriptions. On three datasets, we demonstrate the benefit for visual category learning with zero or few training examples, a critical domain for rare categories or categories defined on the fly.},
  url_pdf = {/publication/jayaraman-2014-zero/jayaraman-2014-zero.pdf},
  url_project = {http://vision.cs.utexas.edu/projects/unreliableAttr/},
  url_code = {https://github.com/dineshj1/UnreliableAttributes},
  url = {https://arxiv.org/abs/1409.4327},
}
@article{jayaraman2012objective, title= {Objective quality assessment of multiply distorted images}, author= {{Jayaraman}, {Dinesh} and Mittal, Anish and Moorthy, Anush K and Bovik, Alan C}, journal= {ASILOMAR Signals, Systems and Computers}, year= {2012},
  abstract = {Subjective studies have been conducted in the past to obtain human judgments of visual quality on distorted images in order, among other things, to benchmark objective image quality assessment (lQA) algorithms. Existing subjective studies primarily have records of human ratings on images that were corrupted by only one of many possible distortions. However, the majority of images that are available for consumption are corrupted by multiple distortions. Towards broadening the corpora of records of human responses to visual distortions, we recently conducted a study on two types of multiply distorted images to obtain human judgments of the visual quality of such images. Further, we compared the performance of several existing objective image quality measures on the new database and analyze the effects of multiple distortions on commonly used quality-determinant features and on human ratings.},
  url_pdf = {/publication/jayaraman-2012-objective/jayaraman-2012-objective.pdf},
  url_project = {http://live.ece.utexas.edu/research/quality/live_multidistortedimage.html},
}