def eval_func(chromosome):
conf = LoadSystemConfiguration()
if not initialPopulationSetted:
setInitialPopulation(gaEngine)
prop = DTIndividualPropertyVanillaEvolutive()
if int(conf.getProperty("Concatenate walk cycles?")):
embryo = DTIndividualGeneticMatrixWalk(chromosomeToLucyGeneticMatrix(chromosome))
else:
embryo = DTIndividualGeneticMatrix(chromosomeToLucyGeneticMatrix(chromosome))
precycleFile = os.getcwd()+"/mocap/cmu_mocap/xml/util/walk_precycle.xml"
preCycleEmbryo = DTIndividualGeneticTimeSerieFile(precycleFile)
preCycleEmbryo.concatenate(embryo)
newEmbryo = preCycleEmbryo
#embryoLength = newEmbryo.getLength()
individual = Individual(prop, newEmbryo)
#individual.setLength(embryoLength)
fitness = individual.execute() #return the fitness resulting from the simulator execution
if int(conf.getProperty("re-evaluate fittest?"))==True:
if fitness > max_score: #is really a better fitness ?
candidateFitness = fitness
fitness = individual.execute()
print "candidateFitness: ", candidateFitness, "fitness: ", fitness
while abs(candidateFitness-fitness) > 0.01:
candidateFitness=fitness
fitness = individual.execute()
print "candidateFitness: ", candidateFitness, "fitness: ", fitness
#the candidateFitness was validated!
fitness = candidateFitness
return fitness
def __init__(self):
conf = LoadSystemConfiguration()
self.geneticMatrix = []
self.cyclesQty = int(conf.getProperty("Concatenate walk cycles?"))
self.robotConfig = LoadRobotConfiguration()
self.jointNameIDMapping = {}
jointIDCounter = 0
for j in self.robotConfig.getJointsName():
self.jointNameIDMapping[jointIDCounter] = j
jointIDCounter += 1
def __init__(self, file):
self.parser = BvhImport(file)
##dtCycleProp = DTWalkPreCyclePropertyLR(file)
##dtCycleProp = DTWalkCycleStartingLeftFootProperty(file)
#dtCycleProp = DTWalkPreCyclePropertyRL(file)
dtCycleProp = DTWalkCycleStartingRightFootProperty(file)
self.start = dtCycleProp.getIndividualStart()
self.end = dtCycleProp.getIndividualEnd()
self.direction = dtCycleProp.getMoveDirection()
conf = LoadSystemConfiguration()
self.skipping = int(conf.getProperty("number of frames to skip"))
def storeExperimentGAparameters():
conf = LoadSystemConfiguration()
file = open(os.path.join(experimentDir,"info.txt"),"w")
file.write("initialPopulationSize = " + conf.getProperty("Population size") + "\n")
file.write("generations = " + conf.getProperty("Number of generations") + "\n")
file.write("genome.crossover = " + conf.getProperty("Crossover operator") + "\n")
file.write("genome.mutator = " + conf.getProperty("Mutator operator") + "\n")
file.write("MutationRate = " + conf.getProperty("MutationRate") + "\n")
file.write("selector = " + conf.getProperty("Selection operator") + "\n")
file.write("CrossoverRate = " + conf.getProperty("CrossoverRate") + "\n")
file.write("ElitismReplacement percentage = " + conf.getProperty("Elitism replacement percentage") + "\n")
file.close()
def storeExperimentGAparameters():
conf = LoadSystemConfiguration()
file = open(os.path.join(experimentDir,"info.txt"),"w")
file.write("initialPopulationSize = " + conf.getProperty("Population size") + "\n")
file.write("generations = " + conf.getProperty("Number of generations") + "\n")
file.write("genome.crossover = " + conf.getProperty("Crossover operator") + "\n")
file.write("genome.mutator = " + conf.getProperty("Mutator operator") + "\n")
file.write("MutationRate = " + conf.getProperty("MutationRate") + "\n")
file.write("selector = " + conf.getProperty("Selection operator") + "\n")
file.write("CrossoverRate = " + conf.getProperty("CrossoverRate") + "\n")
file.write("ElitismReplacement percentage = " + conf.getProperty("Elitism replacement percentage") + "\n")
file.write("Concatenate walk cycles = " + conf.getProperty("Concatenate walk cycles?") + "\n")
file.write("concatenate external cycle file = " + conf.getProperty("concatenate external cycle file?") + "\n")
file.write("Convergence criteria enable? = " + conf.getProperty("Convergence criteria enable?") + "\n")
file.write("Vrep robot model scene = " + conf.getFile("Lucy vrep model") + "\n")
file.close()
def run_main():
conf = LoadSystemConfiguration()
initialPopulationSize = int(conf.getProperty("Population size"))
generations = int(conf.getProperty("Number of generations"))
# Genome instance
framesQty = sysConstants.GENOMA_MAX_LENGTH
genome = G2DList.G2DList(framesQty, 18)
genome.setParams(rangemin=0, rangemax=360)
genome.setParams(gauss_sigma=sysConstants.MUTATION_SIGMA, gauss_mu=0)
# The evaluator function (objective function)
genome.evaluator.set(eval_func)
if conf.getProperty("Crossover operator") == "crossovers.G2DListCrossoverSingleNearHPoint":
genome.crossover.set(crossovers.G2DListCrossoverSingleNearHPoint)
elif conf.getProperty("Crossover operator") == "Crossovers.G2DListCrossoverSingleHPoint":
genome.crossover.set(Crossovers.G2DListCrossoverSingleHPoint)
#genome.crossover.set(crossovers.G2DListCrossoverSingleNearHPointImprove)
#genome.crossover.set(crossovers.G2DListCrossoverSingleHPoint)
# Genetic Algorithm Instance
ga = GSimpleGA.GSimpleGA(genome)
ga.setGenerations(generations)
ga.setPopulationSize(initialPopulationSize)
#genome.mutator.set(Mutators.G2DListMutatorIntegerRange)
if conf.getProperty("Mutator operator") == "mutators.G2DListMutatorRealGaussianSpline":
genome.mutator.set(mutators.G2DListMutatorRealGaussianSpline)
elif conf.getProperty("Mutator operator") == "Mutators.G2DListMutatorRealGaussianGradient":
genome.mutator.set(Mutators.G2DListMutatorRealGaussianGradient)
ga.setMutationRate(float(conf.getProperty("MutationRate")))
if conf.getProperty("Selection operator") == "Selectors.GRankSelector" :
ga.selector.set(Selectors.GRankSelector)
elif conf.getProperty("Selection operator") == "Selectors.GTournamentSelector" :
ga.selector.set(Selectors.GTournamentSelector)
elif conf.getProperty("Selection operator") == "Selectors.GRouletteWheel" :
ga.selector.set(Selectors.GRouletteWheel)
elif conf.getProperty("Selection operator") == "Selectors.GUniformSelector" :
ga.selector.set(Selectors.GUniformSelector)
'''For crossover probability, maybe it is the ratio of next generation population born by crossover operation.
While the rest of population...maybe by previous selection or you can define it as best fit survivors'''
ga.setCrossoverRate(float(conf.getProperty("CrossoverRate")))
elitism = float(conf.getProperty("Elitism replacement percentage")) > 0
ga.setElitism(True)
'''Set the number of best individuals to copy to the next generation on the elitism'''
if elitism:
numberIndividualsForNextGen = int(initialPopulationSize*float(conf.getProperty("Elitism replacement percentage")))
ga.setElitismReplacement(numberIndividualsForNextGen)
if int(conf.getProperty("Convergence criteria enable?"))==True:
ga.terminationCriteria.set(ConvergenceCriteria)
# Create DB Adapter and set as adapter
sqlite_adapter = DBAdapters.DBSQLite(identify="Lucy walk", resetDB=True)
ga.setDBAdapter(sqlite_adapter)
#callback to persist best individual of each generation
ga.stepCallback.set(generationCallback)
#keep a reference to the genetic algorithm engine
global gaEngine
gaEngine = ga
# Do the evolution, with stats dump
# frequency of every generation
ga.evolve(freq_stats=0)
# Best individual
best = ga.bestIndividual()
score = best.getRawScore()
timestr = time.strftime("%Y%m%d-%H%M%S")
filename = str(score) + "-" + timestr + "-" + str(generations) + ".xml"
prop = DTIndividualPropertyVanilla()
bestIndividual = Individual(prop, DTIndividualGeneticMatrix(chromosomeToLucyGeneticMatrix(best)))
bestIndividual.persist(os.path.join(experimentDir,filename))
#store all the final population, not only the fitest
population = ga.getPopulation()
popSize = len(population)
for pos in range(popSize):
individual = Individual(prop, DTIndividualGeneticMatrix(chromosomeToLucyGeneticMatrix(population[pos])))
timestr = time.strftime("%Y%m%d-%H%M%S")
filename = "final-" + str(pos) + "-" + timestr + ".xml"
individual.persist(os.path.join(experimentDir, filename))
#ga.getDBAdapter().commit()
shutil.copy2('pyevolve.db', experimentDir)
shutil.copy2(conf.getProperty("System Log"), experimentDir)
os.system("pyevolve_graph.py -i \"Lucy walk\" -3 -o gene_pool/experiment_img/" + experimentTime + " -e png")
#do the stats
print ga.getStatistics()
class Individual:
def __init__(self, idividualProperty, individualGeneticMaterial):
self.property = idividualProperty
self.fitness = 0
self.robotConfig = LoadRobotConfiguration()
self.configuration = LoadSystemConfiguration()
self.genomeMatrix = individualGeneticMaterial.getGeneticMatrix()
self.poseSize = len(self.genomeMatrix)
self.genomeMatrixJointNameIDMapping = {}
self.sysConf = LoadSystemConfiguration()
i = 0
for jointName in self.robotConfig.getJointsName():
self.genomeMatrixJointNameIDMapping[jointName] = i
i = i + 1
dontSupportedJoints = self.configuration.getVrepNotImplementedBioloidJoints()
self.robotImplementedJoints = []
robotJoints = self.robotConfig.getJointsName()
for joint in robotJoints:
if joint not in dontSupportedJoints:
self.robotImplementedJoints.append(joint)
# is important to use only supported joints to avoid errors obtaining the handler of a joint that doesn't exists
for i in xrange(self.poseSize):
for joint in self.robotImplementedJoints:
# print "i: ", i, "j: ", joint
value = self.genomeMatrix[i][self.genomeMatrixJointNameIDMapping[joint]] + self.property.getPoseFix(
joint
) # TODO this can be a problem for the physical robot
self.genomeMatrix[i][self.genomeMatrixJointNameIDMapping[joint]] = value
def stopLucy(self):
self.lucy.stopLucy()
def execute(self):
# create the corresponding instance of Lucy, depending if it's real or simulated.
if int(self.sysConf.getProperty("Lucy simulated?")) == 1:
self.lucy = SimulatedLucy(int(self.configuration.getProperty("Lucy render enable")))
else:
self.lucy = PhysicalLucy()
poseExecute = {}
i = 0
while self.lucy.isLucyUp() and i < self.poseSize:
for joint in self.robotConfig.getJointsName():
if not (self.property.avoidJoint(joint)):
value = self.genomeMatrix[i][self.genomeMatrixJointNameIDMapping[joint]]
poseExecute[joint] = value
i = i + 1
self.lucy.executePose(Pose(poseExecute))
self.lucy.stopLucy() # this function also updates time and distance
if i < self.poseSize:
self.fitness = self.lucy.getFitness()
else:
endFrameExecuted = True
self.fitness = self.lucy.getFitness(endFrameExecuted)
print "fitness: ", self.fitness
return self.fitness
def getPoseQty(self):
return self.lp.getFrameQty()
def getPose(self, poseNumber):
return self.lp.getFramePose(poseNumber)
def getMostSimilarPose(self, pose):
diff = MAX_INT
moreSimilarPose = self.getPose(1)
for i in xrange(self.getPoseQty()):
myPose = getPose(i)
newDiff = pose.diff(myPose)
if newDiff < diff:
diff = newDiff
moreSimilarPose = myPose
return moreSimilarPose
def getGenomeMatrix(self):
return self.genomeMatrix
def setGenomeMatrix(self, geneMatrix):
self.genomeMatrix = geneMatrix
self.poseSize = len(geneMatrix)
def persist(self, file):
root = ET.Element("root")
lucy = ET.SubElement(root, "Lucy")
for frameIt in xrange(self.poseSize):
frame = ET.SubElement(lucy, "frame")
frame.set("number", str(frameIt))
for joint in self.robotImplementedJoints:
xmlJoint = ET.SubElement(frame, joint)
joint_id = self.robotConfig.loadJointId(joint)
pos = self.genomeMatrix[frameIt][self.genomeMatrixJointNameIDMapping[joint]]
xmlJointAngle = xmlJoint.set("angle", str(pos))
tree = ET.ElementTree(root)
tree.write(file)
class Individual:
def __init__(self, idividualProperty, individualGeneticMaterial, modelReposeValue=DTModelVrepReda()):
self.property = idividualProperty
self.modelReposeValue = modelReposeValue
self.fitness = 0
self.robotConfig = LoadRobotConfiguration()
self.configuration = LoadSystemConfiguration()
self.genomeMatrix = individualGeneticMaterial.getGeneticMatrix()
self.length = individualGeneticMaterial.getLength()
self.genomeMatrixJointNameIDMapping = {}
self.sysConf = LoadSystemConfiguration()
self.individualGeneticMaterial = individualGeneticMaterial
self.moveArmFirstTime = True
self.precycleLength = 0
self.cycleLength = 0 #Not used yet
i=0
for jointName in self.robotConfig.getJointsName():
self.genomeMatrixJointNameIDMapping[jointName]=i
i=i+1
dontSupportedJoints = self.configuration.getVrepNotImplementedBioloidJoints()
self.robotImplementedJoints = []
robotJoints = self.robotConfig.getJointsName()
for joint in robotJoints:
if joint not in dontSupportedJoints:
self.robotImplementedJoints.append(joint)
#is important to use only supported joints to avoid errors obtaining the handler of a joint that doesn't exists
for i in xrange(self.length):
for joint in self.robotImplementedJoints:
#print "i: ", i, "j: ", joint
if self.property.avoidJoint(joint):
value = modelReposeValue.getReposeValue(joint)
else:
value = self.genomeMatrix[i][self.genomeMatrixJointNameIDMapping[joint]] + self.property.getPoseFix(joint) #TODO this can be a problem for the physical robot
self.genomeMatrix[i][self.genomeMatrixJointNameIDMapping[joint]]=value
def stopLucy(self):
self.lucy.stopLucy()
def execute(self):
#create the corresponding instance of Lucy, depending if it's real or simulated.
if int(self.sysConf.getProperty("Lucy simulated?"))==1:
self.lucy = SimulatedLucy(int(self.configuration.getProperty("Lucy render enable")))
else:
self.lucy = PhysicalLucy()
poseExecute={}
i=0
while (self.lucy.isLucyUp() and i < self.length):
for joint in self.robotConfig.getJointsName():
if not(self.property.avoidJoint(joint)):
value = self.genomeMatrix[i][self.genomeMatrixJointNameIDMapping[joint]]
poseExecute[joint] = value
i = i + self.lucy.getPosesExecutedByStepQty()
self.lucy.executePose(Pose(poseExecute))
'''if self.precycleLength > 0 and i == self.precycleLength - 1:
self.lucy.moveHelperArm()
if self.precycleLength > 0 and i > self.precycleLength - 1:
if (i - self.precycleLength) > 0 and (i - self.precycleLength) % self.cycleLength == 0:
self.lucy.moveHelperArm()'''
startingCyclePose = self.getPrecycleLength()
executionConcatenationGap = self.individualGeneticMaterial.getConcatenationGap(startingCyclePose)
self.fitness = self.lucy.getFitness(self.length, executionConcatenationGap)
self.lucy.stopLucy() #this function also updates time and distance
return self.fitness
def getGenomeMatrix(self):
return self.genomeMatrix
def setGenomeMatrix(self, geneMatrix):
self.genomeMatrix = geneMatrix
dtGenome = DTGenomeFunctions()
self.length = dtGenome.individualLength(geneMatrix)
def persist(self,file):
root = ET.Element("root")
lucy = ET.SubElement(root, "Lucy")
for frameIt in xrange(self.length):
frame = ET.SubElement(lucy, "frame")
frame.set("number" , str(frameIt))
for joint in self.robotImplementedJoints: #TODO because the notImplementedJoints doesn't participate in the simulation and fitness evaluation, they are not stored
xmlJoint = ET.SubElement(frame, joint)
joint_id = self.robotConfig.loadJointId(joint)
pos = self.genomeMatrix[frameIt][self.genomeMatrixJointNameIDMapping[joint]]
xmlJointAngle = xmlJoint.set("angle" , str(pos))
tree = ET.ElementTree(root)
tree.write(file)
def getJointMatrixIDFromName(self, jointName):
return self.genomeMatrixJointNameIDMapping[jointName]
def setLength(self, length):
self.length = length
def setPrecycleLength(self, precycleLength):
self.precycleLength = precycleLength
def setCycleLength(self, cycleLength):
self.cycleLength = cycleLength
def getLength(self):
return self.length
def getPrecycleLength(self):
return self.precycleLength
def getCycleLength(self):
return self.cycleLength
