jangirrishabh

"""
Scene file for cloth on table.
To check deformations of the cloth under Attach Constraint and Fixed Constraint
Current script uses only Attach constraint at the 2 red balls to grasp the cloth
To add Fixed constraint uncomment line 62

Note: The red balls are supposed to move together in simulation and thus the cloth must follow them with minimum deformations
at the grasped vertices
"""

jangirrishabh

class workspace (Sofa.PythonScriptController):

    def __init__(self, node, commandLineArguments) : 
        self.commandLineArguments = commandLineArguments
        print "Command line arguments for python : "+str(commandLineArguments)
        self.node = node
        self.createGraph(node)
        return None;

    def createGraph(self,rootNode):

jangirrishabh

<?xml version="1.0" ?>
<!-- Testing ball collision problem from hugo -->
<Node name="root" dt="0.05" gravity="0 -9.810 0">
    <VisualStyle displayFlags="hideBehaviorModels hideCollisionModels hideMappings hideForceFields showVisualModels hideInteractionForceFields" />
    <RequiredPlugin name='SofaMiscCollision'/>
    <RequiredPlugin name="SofaPython" pluginName="SofaPython" />

    <FreeMotionAnimationLoop solveVelocityConstraintFirst="0" />
    <LCPConstraintSolver maxIt="1000" tolerance="1e-6" initial_guess="false" build_lcp="true" multi_grid="false" printLog="0" mu="0.000000000001"/>

jangirrishabh

<Node name="root" dt="0.05" gravity="0 -9.810 0">
    <VisualStyle displayFlags="hideBehaviorModels hideCollisionModels hideMappings hideForceFields showVisualModels hideInteractionForceFields" />
    <RequiredPlugin name='SofaMiscCollision'/>
    <RequiredPlugin name="SofaPython" pluginName="SofaPython" />

    <FreeMotionAnimationLoop solveVelocityConstraintFirst="0" />
    <LCPConstraintSolver maxIt="1000" tolerance="1e-6" initial_guess="false" build_lcp="true" multi_grid="false" printLog="0" mu="1.9"/>

    <CollisionPipeline depth="6" verbose="0" draw="0" />
    <BruteForceDetection name="N2" />

jangirrishabh

<Node name="root" dt="0.05" gravity="0 -9.810 0">
    <VisualStyle displayFlags="hideBehaviorModels hideCollisionModels hideMappings hideForceFields showVisualModels" />
    <RequiredPlugin name='SofaMiscCollision'/>
    <LCPConstraintSolver maxIt="1000" tolerance="1e-6" initial_guess="false" build_lcp="true" multi_grid="false" printLog="0" mu="0.9"/>
    <FreeMotionAnimationLoop solveVelocityConstraintFirst="1" />
    <CollisionPipeline depth="15" verbose="0" draw="0" />
    <BruteForceDetection name="N2" />
    <LocalMinDistance name="Proximity" alarmDistance="0.3" contactDistance="0.1" useLMDFilters="0" />
    <CollisionResponse name="Response" response="FrictionContact" />
    <DiscreteIntersection/>

jangirrishabh

import Sofa

class KeyboardControl(Sofa.PythonScriptController):
	# called once graph is created, to init some stuff...
	def initGraph(self,node):
		print 'initGraph called (python side)'
		self.MechanicalState = node.getObject('DOFs')
		gravity = node.findData('gravity').value
		print gravity
		self.rootNode = node.getRoot()

jangirrishabh

    def optimalWeightFinder(self):
        i = 1
        while True:
            W = self.optimization() # optimize to find new weights in the list of policies
            print ("weights ::", W )
            print ("the distances  ::", self.policiesFE.keys())
            self.currentT = self.policyListUpdater(W, i)
            print ("Current distance (t) is:: ", self.currentT )
            if self.currentT <= self.epsilon: # terminate if the point reached close enough
                break

jangirrishabh

class irlAgent:
    def __init__(self, randomFE, expertFE, epsilon, num_states, num_frames, behavior):
        self.randomPolicy = randomFE
        self.expertPolicy = expertFE
        self.num_states = num_states
        self.num_frames = num_frames
        self.behavior = behavior
        self.epsilon = epsilon # termination when t < 0.1
        self.randomT = np.linalg.norm(np.asarray(self.expertPolicy)-np.asarray(self.randomPolicy)) #norm of the diff in expert and random
        self.policiesFE = {self.randomT:self.randomPolicy} # storing the policies and their respective t values in a dictionary

jangirrishabh

    def getRLAgentFE(self, W, i): #get the feature expectations of a new poliicy using RL agent
        IRL_helper(W, self.behavior, self.num_frames, i) # train the agent and save the model in a file used below
        saved_model = 'saved-models_'+self.behavior+str(i)+'/164-150-100-50000-'+str(self.num_frames)+'.h5' # use the saved model to get the FE
        model = neural_net(self.num_states, [164, 150], saved_model)
        return  play(model, W)#return feature expectations by executing the learned policy

jangirrishabh

    def policyListUpdater(self, W, i):  #add the policyFE list and differences
        tempFE = self.getRLAgentFE(W, i) # get feature expectations of a new policy respective to the input weights
        hyperDistance = np.abs(np.dot(W, np.asarray(self.expertPolicy)-np.asarray(tempFE))) #hyperdistance = t
        self.policiesFE[hyperDistance] = tempFE
        return hyperDistance # t = (weights.tanspose)*(expert-newPolicy)