def main():
# MDP Env
xBoundary = [0, 600]
yBoundary = [0, 600]
xSwamp = [300, 400]
ySwamp = [300, 400]
swamp = [[[300, 400], [300, 400]]]
noise = [50, 50]
stayInBoundaryByReflectVelocity = StayInBoundaryByReflectVelocity(
xBoundary, yBoundary)
transitionWithNoise = TransitionWithNoise(noise)
minDistance = 50
target = [600, 600]
isTerminal = IsTerminal(minDistance, target)
isInSwamp = IsInSwamp(swamp)
singleAgentTransit = MovingAgentTransitionInSwampWorld(
transitionWithNoise, stayInBoundaryByReflectVelocity, isTerminal)
transitionFunctionPack = [singleAgentTransit, static]
multiAgentTransition = MultiAgentTransitionInGeneral(
transitionFunctionPack)
twoAgentTransit = MultiAgentTransitionInSwampWorld(multiAgentTransition,
target)
numOfAgent = 2
xBoundaryReset = [500, 600]
yBoundaryReset = [0, 100]
resetState = Reset([0, 0], [0, 0], numOfAgent, target)
actionSpace = [(100, 0), (-100, 0), (0, 100), (0, -100)]
#k = np.random.choice(actionSpace)
#print(k)
actionCost = -1
swampPenalty = -100
terminalReward = 1000
rewardFunction = RewardFunction(actionCost, terminalReward, swampPenalty,
isTerminal, isInSwamp)
maxRunningSteps = 100
oneStepSampleTrajectory = OneStepSampleTrajectory(twoAgentTransit,
rewardFunction)
sampleTrajecoty = SampleTrajectory(maxRunningSteps, isTerminal, resetState,
oneStepSampleTrajectory)
randomPolicy = RandomPolicy(actionSpace)
actionDistribution = randomPolicy()
#numSimulation, selectAction, selectNextState, expand, estimateValue, backup, outputDistribution
numSimulation = 5
cInit = 100
cBase = 1
scoreChild = ScoreChild(cInit, cBase)
selectAction = SelectAction(scoreChild)
selectNextState = SelectNextState(selectAction)
uniformActionPrior = {action: 1 / 4 for action in actionSpace}
getActionPrior = lambda state: uniformActionPrior
initializeChildren = InitializeChildren(actionSpace, twoAgentTransit,
getActionPrior)
expand = Expand(isTerminal, initializeChildren)
alpha = 0
C = 3
pWidening = PWidening(alpha, C)
expandNewState = ExpandNextState(twoAgentTransit, pWidening)
rolloutPolicy = lambda state: random.choice(actionSpace)
rolloutHeuristic = lambda state: 0 #reward return sometimes grab nothing.
maxRolloutStep = 10
estimateValue = RollOut(rolloutPolicy, maxRolloutStep, twoAgentTransit,
rewardFunction, isTerminal, rolloutHeuristic)
mctsSelectAction = MCTS(numSimulation, selectAction, selectNextState,
expand, expandNewState, estimateValue, backup,
establishPlainActionDist)
#sampleAction = SampleFromDistribution(actionDictionary)
def sampleAction(state):
actionDist = mctsSelectAction(state)
action = maxFromDistribution(actionDist)
return action
trajectories = [sampleTrajecoty(sampleAction) for _ in range(1)]
#print(findCumulativeReward(trajectories))
print(trajectories)
DIRNAME = os.path.dirname(__file__)
trajectoryDirectory = os.path.join(DIRNAME, '..', '..', 'data',
'evaluateObstacle2', 'trajectories')
if not os.path.exists(trajectoryDirectory):
os.makedirs(trajectoryDirectory)
# generate demo image
screenWidth = 600
screenHeight = 600
screen = pg.display.set_mode((screenWidth, screenHeight))
screenColor = THECOLORS['black']
xBoundary = [0, 600]
yBoundary = [0, 600]
lineColor = THECOLORS['white']
lineWidth = 4
xSwamp = [300, 400]
ySwamp = [300, 400]
drawBackground = DrawBackground(screen, screenColor, xBoundary, yBoundary,
lineColor, lineWidth, xSwamp, ySwamp)
fps = 40
circleColorSpace = np.array([[0, 0, 255], [0, 255, 255]])
circleSize = 10
positionIndex = [0, 1]
saveImage = True
imageSavePath = os.path.join(trajectoryDirectory, 'picMovingSheep')
if not os.path.exists(imageSavePath):
os.makedirs(imageSavePath)
trajectoryParameters = 'obstacle'
imageFolderName = str(trajectoryParameters)
saveImageDir = os.path.join(os.path.join(imageSavePath, imageFolderName))
if not os.path.exists(saveImageDir):
os.makedirs(saveImageDir)
agentIdsToDraw = list(range(2))
drawState = DrawState(fps, screen, circleColorSpace, circleSize,
agentIdsToDraw, positionIndex, saveImage,
saveImageDir, drawBackground)
numFramesToInterpolate = 3
interpolateState = InterpolateState(numFramesToInterpolate,
twoAgentTransit)
stateIndexInTimeStep = 0
actionIndexInTimeStep = 1
chaseTrial = ChaseTrialWithTraj(stateIndexInTimeStep, drawState,
interpolateState, actionIndexInTimeStep)
[chaseTrial(trajectory) for trajectory in trajectories]
pg.quit()
def main():
# MDP Env
xBoundary = [0, 600]
yBoundary = [0, 600]
xSwamp = [300, 400]
ySwamp = [300, 400]
swamp = [[[300,400],[300,400]]]
noise = [1, 1]
stayInBoundaryByReflectVelocity = StayInBoundaryByReflectVelocity(xBoundary, yBoundary)
transitionWithNoise = TransitionWithNoise(noise)
minDistance = 50
target = [200, 200]
isTerminal = IsTerminal(minDistance, target)
isInSwamp = IsInSwamp(swamp)
singleAgentTransit = MovingAgentTransitionInSwampWorld(transitionWithNoise, stayInBoundaryByReflectVelocity, isTerminal)
transitionFunctionPack = [singleAgentTransit, static]
multiAgentTransition = MultiAgentTransitionInGeneral(transitionFunctionPack)
twoAgentTransit = MultiAgentTransitionInSwampWorld(multiAgentTransition, target)
numOfAgent = 2
xBoundaryReset = [500, 600]
yBoundaryReset = [0, 100]
resetState = Reset(xBoundaryReset, yBoundaryReset, numOfAgent, target)
actionSpace = [[10, 0], [-10, 0], [-10, -10], [10, 10], [0, 10], [0, -10], [-10, 10], [10, -10]]
actionCost = -1
swampPenalty = -10
terminalReward = 10
rewardFunction = RewardFunction(actionCost, terminalReward, swampPenalty, isTerminal, isInSwamp)
maxRunningSteps = 100
oneStepSampleTrajectory = OneStepSampleTrajectory(twoAgentTransit, rewardFunction)
sampleTrajecoty = SampleTrajectory(maxRunningSteps, isTerminal, resetState, oneStepSampleTrajectory)
randomPolicy = RandomPolicy(actionSpace)
actionDistribution = randomPolicy()
sampleAction = SampleFromDistribution(actionDistribution)
trajectories = [sampleTrajecoty(sampleAction) for _ in range(10)]
DIRNAME = os.path.dirname(__file__)
trajectoryDirectory = os.path.join(DIRNAME, '..', '..', 'data', 'evaluateObstacle',
'trajectories')
if not os.path.exists(trajectoryDirectory):
os.makedirs(trajectoryDirectory)
# generate demo image
screenWidth = 600
screenHeight = 600
screen = pg.display.set_mode((screenWidth, screenHeight))
screenColor = THECOLORS['black']
xBoundary = [0, 600]
yBoundary = [0, 600]
lineColor = THECOLORS['white']
lineWidth = 4
xSwamp=[300,400]
ySwamp=[300,400]
drawBackground = DrawBackground(screen, screenColor, xBoundary, yBoundary, lineColor, lineWidth, xSwamp, ySwamp)
fps=40
circleColorSpace = np.array([[0, 0, 255], [0, 255, 255] ])
circleSize = 10
positionIndex = [0, 1]
saveImage = True
imageSavePath = os.path.join(trajectoryDirectory, 'picMovingSheep')
if not os.path.exists(imageSavePath):
os.makedirs(imageSavePath)
trajectoryParameters = 'obstacle'
imageFolderName = str(trajectoryParameters)
saveImageDir = os.path.join(os.path.join(imageSavePath, imageFolderName))
if not os.path.exists(saveImageDir):
os.makedirs(saveImageDir)
agentIdsToDraw = list(range(2))
drawState = DrawState(fps, screen, circleColorSpace, circleSize, agentIdsToDraw, positionIndex,
saveImage, saveImageDir, drawBackground)
numFramesToInterpolate = 3
interpolateState = InterpolateState(numFramesToInterpolate, twoAgentTransit)
stateIndexInTimeStep = 0
actionIndexInTimeStep = 1
chaseTrial = ChaseTrialWithTraj(stateIndexInTimeStep, drawState, interpolateState, actionIndexInTimeStep)
[chaseTrial(trajectory) for trajectory in trajectories]
pg.quit()
def main():
DIRNAME = os.path.dirname(__file__)
trajectoryDirectory = os.path.join(DIRNAME, '..', '..', 'data', 'evaluateHierarchyPlanning',
'trajectories')
if not os.path.exists(trajectoryDirectory):
os.makedirs(trajectoryDirectory)
NNNumSimulations = 250
maxRunningSteps = 52
softParameterInPlanningForSheep = 2.5
softParameterInPlanning = 2.5
hierarchy = 0
trajectoryFixedParameters = {'sheepPolicySoft': softParameterInPlanningForSheep, 'wolfPolicySoft': softParameterInPlanning,
'maxRunningSteps': maxRunningSteps, 'hierarchy': hierarchy, 'NNNumSimulations':NNNumSimulations}
trajectoryExtension = '.pickle'
getTrajectorySavePath = GetSavePath(trajectoryDirectory, trajectoryExtension, trajectoryFixedParameters)
# Compute Statistics on the Trajectories
loadTrajectories = LoadTrajectories(getTrajectorySavePath, loadFromPickle)
numWolves = 2
numSheep = 2
valuePriorEndTime = -100
valuePriorSoftMaxBeta = 0.0
trajectoryParameters = {'numWolves': numWolves, 'numSheep': numSheep, 'valuePriorEndTime': valuePriorEndTime,
'valuePriorSoftMaxBeta': valuePriorSoftMaxBeta}
wolfType = 'sharedReward'
#trajectoryParameters.update({'wolfType': wolfType})
trajectories = loadTrajectories(trajectoryParameters)
# generate demo image
screenWidth = 600
screenHeight = 600
screen = pg.display.set_mode((screenWidth, screenHeight))
screenColor = THECOLORS['black']
xBoundary = [0, 600]
yBoundary = [0, 600]
lineColor = THECOLORS['white']
lineWidth = 4
drawBackground = DrawBackground(screen, screenColor, xBoundary, yBoundary, lineColor, lineWidth)
FPS = 32
circleColorSpace = [[100, 100, 100]]*numSheep + [[255, 255, 255]] * numWolves
circleSize = 10
positionIndex = [0, 1]
agentIdsToDraw = list(range(numSheep + numWolves))
saveImage = True
imageSavePath = os.path.join(trajectoryDirectory, 'picMovingSheep')
if not os.path.exists(imageSavePath):
os.makedirs(imageSavePath)
imageFolderName = str('forDemo')
saveImageDir = os.path.join(os.path.join(imageSavePath, imageFolderName))
if not os.path.exists(saveImageDir):
os.makedirs(saveImageDir)
goalSpace = list(range(numSheep))
imaginedWeIdsForInferenceSubject = list(range(numSheep, numWolves + numSheep))
softParameter = 1
softFunction = SoftDistribution(softParameter)
updateColorSpaceByPosterior = lambda colorSpace, posterior : updateColorSpace(
colorSpace, [softFunction(individualPosterior) for individualPosterior in posterior], goalSpace, imaginedWeIdsForInferenceSubject)
#updateColorSpaceByPosterior = lambda originalColorSpace, posterior : originalColorSpace
outsideCircleAgentIds = imaginedWeIdsForInferenceSubject
outsideCircleColor = np.array([[255, 0, 0]] * numWolves)
outsideCircleSize = 15
drawCircleOutside = DrawCircleOutside(screen, outsideCircleAgentIds, positionIndex, outsideCircleColor, outsideCircleSize)
drawState = DrawState(FPS, screen, circleColorSpace, circleSize, agentIdsToDraw, positionIndex,
saveImage, saveImageDir, drawBackground, updateColorSpaceByPosterior, drawCircleOutside)
# MDP Env
xBoundary = [0,600]
yBoundary = [0,600]
stayInBoundaryByReflectVelocity = StayInBoundaryByReflectVelocity(xBoundary, yBoundary)
transit = InterpolateOneFrame(stayInBoundaryByReflectVelocity)
numFramesToInterpolate = 7
interpolateState = InterpolateState(numFramesToInterpolate, transit)
stateIndexInTimeStep = 0
actionIndexInTimeStep = 1
posteriorIndexInTimeStep = 4
chaseTrial = ChaseTrialWithTraj(stateIndexInTimeStep, drawState, interpolateState, actionIndexInTimeStep, posteriorIndexInTimeStep)
print(len(trajectories))
lens = [len(trajectory) for trajectory in trajectories]
index = np.argsort(-np.array(lens))
print(index)
print(trajectories[0][1])
[chaseTrial(trajectory) for trajectory in np.array(trajectories)[index[0:10]]]
print([len(trajectory) for trajectory in np.array(trajectories)[index[:]]])
def main():
DIRNAME = os.path.dirname(__file__)
trajectoryDirectory = os.path.join(DIRNAME, '..', '..', 'data',
'evaluateCompeteDetection',
'trajectories')
if not os.path.exists(trajectoryDirectory):
os.makedirs(trajectoryDirectory)
NNNumSimulations = 250
maxRunningSteps = 61
softParameterInPlanningForSheep = 2.0
softParameterInPlanning = 2.0
trajectoryFixedParameters = {
'sheepPolicySoft': softParameterInPlanningForSheep,
'wolfPolicySoft': softParameterInPlanning,
'maxRunningSteps': maxRunningSteps,
'NNNumSimulations': NNNumSimulations
}
trajectoryExtension = '.pickle'
getTrajectorySavePath = GetSavePath(trajectoryDirectory,
trajectoryExtension,
trajectoryFixedParameters)
# Compute Statistics on the Trajectories
loadTrajectories = LoadTrajectories(getTrajectorySavePath, loadFromPickle)
numWolves = 2
numSheep = 1
competePolicy = 'heatseeking'
heatseekingPrecesion = 1.83
otherCompeteRate = 1.0
competeDetectionRate = 0.5
inferenceSoft = 0.05
trajectoryParameters = {
'heatseekingPrecesion': heatseekingPrecesion,
'inferenceSoft': inferenceSoft,
'numWolves': numWolves,
'numSheep': numSheep,
'competePolicy': competePolicy,
'otherCompeteRate': otherCompeteRate,
'competeDetectionRate': competeDetectionRate
}
trajectories = loadTrajectories(trajectoryParameters)
# generate demo image
screenWidth = 600
screenHeight = 600
screen = pg.display.set_mode((screenWidth, screenHeight))
screenColor = THECOLORS['black']
xBoundary = [0, 600]
yBoundary = [0, 600]
lineColor = THECOLORS['white']
lineWidth = 4
drawBackground = DrawBackground(screen, screenColor, xBoundary, yBoundary,
lineColor, lineWidth)
FPS = 24
circleColorSpace = [[0, 255, 0]] * numSheep + [[255, 0, 0]] * numWolves
circleSize = 10
positionIndex = [0, 1]
agentIdsToDraw = list(range(numSheep + numWolves))
#saveImage = False
saveImage = True
imageSavePath = os.path.join(trajectoryDirectory, 'picMovingSheep')
if not os.path.exists(imageSavePath):
os.makedirs(imageSavePath)
imageFolderName = str('forDemo')
saveImageDir = os.path.join(os.path.join(imageSavePath, imageFolderName))
if not os.path.exists(saveImageDir):
os.makedirs(saveImageDir)
goalSpace = list(range(numSheep))
imaginedWeIdsForInferenceSubject = list(
range(numSheep, numWolves + numSheep))
softParameter = 1.1
softFunction = SoftDistribution(softParameter)
selfPosteriorIndex = 0
concernedAgentId = 2
competeIntention = (0, ())
cooperateIntention = (0, tuple(range(numSheep, numSheep + numWolves)))
updateColorSpaceByPosterior = lambda colorSpace, posterior: updateColorSpace(
colorSpace, [
softFunction(individualPosterior)
for individualPosterior in posterior
], selfPosteriorIndex, concernedAgentId, competeIntention,
cooperateIntention)
#updateColorSpaceByPosterior = lambda originalColorSpace, posterior : originalColorSpace
outsideCircleAgentIds = imaginedWeIdsForInferenceSubject
outsideCircleColor = np.array([[0, 0, 0]] * numWolves)
outsideCircleSize = 15
drawCircleOutside = DrawCircleOutside(screen, outsideCircleAgentIds,
positionIndex, outsideCircleColor,
outsideCircleSize)
drawState = DrawState(FPS, screen, circleColorSpace, circleSize,
agentIdsToDraw, positionIndex, saveImage,
saveImageDir, drawBackground,
updateColorSpaceByPosterior, drawCircleOutside)
# MDP Env
xBoundary = [0, 600]
yBoundary = [0, 600]
stayInBoundaryByReflectVelocity = StayInBoundaryByReflectVelocity(
xBoundary, yBoundary)
transit = InterpolateOneFrame(stayInBoundaryByReflectVelocity)
numFramesToInterpolate = 5
interpolateState = InterpolateState(numFramesToInterpolate, transit)
stateIndexInTimeStep = 0
actionIndexInTimeStep = 1
posteriorIndexInTimeStep = 4
chaseTrial = ChaseTrialWithTraj(stateIndexInTimeStep, drawState,
interpolateState, actionIndexInTimeStep,
posteriorIndexInTimeStep)
print(len(trajectories))
lens = [len(trajectory) for trajectory in trajectories]
index = np.argsort(-np.array(lens))
print(index)
print(trajectories[0][1])
#[chaseTrial(trajectory) for trajectory in np.array(trajectories)[index[0:10]]]
[
chaseTrial(trajectory)
for trajectory in np.array(trajectories)[index[9:10]]
]
def main():
manipulatedVariables = OrderedDict()
manipulatedVariables['numOfAgent'] = [2]
levelNames = list(manipulatedVariables.keys())
levelValues = list(manipulatedVariables.values())
modelIndex = pd.MultiIndex.from_product(levelValues, names=levelNames)
toSplitFrame = pd.DataFrame(index=modelIndex)
productedValues = it.product(
*[[(key, value) for value in values]
for key, values in manipulatedVariables.items()])
parametersAllCondtion = [
dict(list(specificValueParameter))
for specificValueParameter in productedValues
]
numTrajectories = 3
sampleTrajectoriesForConditions = SampleTrajectoriesForCoditions(
numTrajectories, composeFowardOneTimeStepWithRandomSubtlety)
trajectoriesMultipleConditions = [
sampleTrajectoriesForConditions(para) for para in parametersAllCondtion
]
visualConditionIndex = 0
trajectoriesToVisualize = trajectoriesMultipleConditions[
visualConditionIndex]
visualize = True
if visualize:
screenWidth = 640
screenHeight = 480
screen = pg.display.set_mode((screenWidth, screenHeight))
screenColor = THECOLORS['black']
xBoundary = [0, 640]
yBoundary = [0, 480]
lineColor = THECOLORS['white']
lineWidth = 4
drawBackground = DrawBackground(screen, screenColor, xBoundary,
yBoundary, lineColor, lineWidth)
numOfAgent = 2
numDistractors = numOfAgent - 2
circleColorSpace = [[0, 255, 0], [255, 0, 0]
] + [[255, 255, 255]] * numDistractors
circleSize = 10
positionIndex = [0, 1]
agentIdsToDraw = list(range(numOfAgent))
saveImage = False
dirPYFile = os.path.dirname(__file__)
imageSavePath = os.path.join(dirPYFile, '..', 'data', 'forDemo')
if not os.path.exists(imageSavePath):
os.makedirs(imageSavePath)
FPS = 30
drawState = DrawState(FPS, screen, circleColorSpace, circleSize,
agentIdsToDraw, positionIndex, saveImage,
imageSavePath, drawBackground)
# MDP Env
xBoundary = [0, 640]
yBoundary = [0, 480]
stayInBoundaryByReflectVelocity = StayInBoundaryByReflectVelocity(
xBoundary, yBoundary)
distanceToVisualDegreeRatio = 20
killzoneRadius = 2.5 * distanceToVisualDegreeRatio
sheepId = 0
wolfId = 1
isTerminal = IsTerminal(sheepId, wolfId, killzoneRadius)
numMDPTimeStepPerSecond = 5 # change direction every 200ms
numFramesToInterpolate = int(
FPS / numMDPTimeStepPerSecond - 1
) # interpolate each MDP timestep to multiple frames; check terminal for each frame
interpolateStateForVisualization = InterpolateStateForVisualization(
numFramesToInterpolate, stayInBoundaryByReflectVelocity,
isTerminal)
stateIndexInTimeStep = 0
actionIndexInTimeStep = 1
nextStateIndexInTimeStep = 2
visualizeTraj = VisualizeTraj(stateIndexInTimeStep,
actionIndexInTimeStep,
nextStateIndexInTimeStep, drawState,
interpolateStateForVisualization)
[visualizeTraj(trajectory) for trajectory in trajectoriesToVisualize]