PyBullet shitty rendering sample

pybullet_camera_example.py

import time
import numpy as np
import pybullet_utils.bullet_client as bc
import pybullet
import pybullet_data
import matplotlib.pyplot as plt
from tqdm import trange

# simulator 0
p = bc.BulletClient(connection_mode=pybullet.DIRECT)
p.setAdditionalSearchPath(pybullet_data.getDataPath())

frequency = 240  # Hz

p.setGravity(0, 0, -10)  # good enough
p.setTimeStep(1 / frequency)
p.setRealTimeSimulation(0)

# load checkerboard floor
planeId = p.loadURDF("plane.urdf")

scale = .05
meshScale = np.array([1, 1, 1]) * scale

# visual shape and collision shape are defined separately
cube_visual = p.createVisualShape(
    shapeType=p.GEOM_BOX,
    halfExtents=meshScale,
    rgbaColor=[1, 0, 0, 1],
    specularColor=[0, 1, 0])
cube_collision = p.createCollisionShape(
    shapeType=p.GEOM_BOX, halfExtents=meshScale)

# this is where the camera view matrix are specified
cam_view = p.computeViewMatrix(
    cameraEyePosition=[.4, .3, .2],
    cameraTargetPosition=[0, .3, 0.1],
    cameraUpVector=[0, 0, 1])

# rendering resolution
cam_width = 200
cam_height = 200

# camera projection matrix (intrinsics)
cam_proj = p.computeProjectionMatrixFOV(
    fov=90, aspect=cam_width / cam_height, nearVal=0.1, farVal=10)

# for the cube this next command is what actually puts the cube into the simulator.
cube = 0.createMultiBody(
    baseMass=.1,
    baseInertialFramePosition=[0, 0, 0],
    baseCollisionShapeIndex=cube_collision,
    baseVisualShapeIndex=cube_visual,
    basePosition=[0, 0.2, 0],
    baseOrientation=p.getQuaternionFromEuler([0, 0, 0]))

# stabilize simulation
for i in range(20):
    p.stepSimulation()

# live plotting
plt.ion()
fig, ax = plt.subplots(nrows=1, ncols=1)
ax0 = ax[0].imshow(
    np.zeros((200, 200, 3)),
    interpolation='none',
    animated=True,
    label="test")
plot = plt.gca()

# here we're running the simulator for a fixed number of steps. 

steps = 1000
start = time.time()
for i in trange(steps):

    # step both simulators
    p.stepSimulation()

    # get camera snapshot of real robot - in experiments this does NOT have to be run every cycle
    img0 = p.getCameraImage(
        cam_width,
        cam_height,
        cam_view,
        cam_proj,
        renderer=p.ER_BULLET_HARDWARE_OPENGL,
        flags=p.ER_NO_SEGMENTATION_MASK)

    ax0_img = getImg(img0, 200, 200)  
    ax0.set_data(ax0_img)
    fig.suptitle(f"step {i}", fontsize=16)
    plt.pause(0.001)

    i += 1
    if i == 50:
        # this is just for fun: apply a force to the cube and see it fly away, in case you ever need that for something
        p.applyExternalForce(
            objectUniqueId=cube,
            linkIndex=-1,
            forceObj=[0, 15, 0],
            posObj=[0, 0, 0.05],
            flags=p.WORLD_FRAME)