def __init__(self, thymioController, debug, experiment_name):
threading.Thread.__init__(self)
config = ConfigParser(cl.CONFIG_PATH)
self.__address = config.address
self.__port = config.port
self.__thymioController = thymioController
self.__tcPerformedAction = threading.Condition()
self.__tcPA = False
self.__msgSenders = dict()
self.__msgReceivers = dict()
self.__stopSockets = list()
self.__stopped = False
self.__previous_motor_speed = [0, 0]
# simulation logging file
self.__simLogger = logging.getLogger('simulationLogger')
logLevel = logging.INFO
if debug:
logLevel = logging.DEBUG
self.__simLogger.setLevel(logLevel)
self.__experiment_name = experiment_name
outputDir = os.path.join(cl.OUTPUT_PATH, self.__experiment_name)
mkdir_p(outputDir)
# self.__nextSimFilename = getNextIDPath(SIM_LOG_PATH) + '_' + time.strftime("%Y-%m-%d_%H-%M-%S")
# simHandler = logging.FileHandler(os.path.join(SIM_LOG_PATH, self.__nextSimFilename + '_sim_debug.log'))
self.__simulationLogFile = os.path.join(outputDir, self.__experiment_name + '_sim_debug.log')
self.__simulationOutputFile = os.path.join(outputDir, self.__experiment_name + '_out.txt')
self.__simulationWeightOutputFile = os.path.join(outputDir, self.__experiment_name + '_weight_out.txt')
self.__simulationTempFile = os.path.join(outputDir, self.__experiment_name + '_temp.txt')
simHandler = logging.FileHandler(self.__simulationLogFile)
simHandler.setFormatter(cl.FORMATTER)
self.__simLogger.addHandler(simHandler)
self.__threadCamera = cameraVision(False, self.__simLogger)
self.__inbox = Inbox(self.__simLogger)
for bot in config.bots:
address = bot["address"]
self.__msgSenders[address] = MessageSender(address, bot["port"], self.__simLogger)
self.__msgReceivers[address] = MessageReceiver(address, self.__inbox, self.__simLogger)
self.__connListener = ConnectionsListener(self.__address, self.__port, self.__msgReceivers, self.__simLogger)
def __init__(self, thymioController, debug, experiment_name):
threading.Thread.__init__(self)
config = ConfigParser(cl.CONFIG_PATH)
self.__address = config.address
self.__port = config.port
self.__thymioController = thymioController
self.__tcPerformedAction = threading.Condition()
self.__tcPA = False
self.__msgSenders = dict()
self.__msgReceivers = dict()
self.__stopSockets = list()
self.__stopped = False
self.__previous_motor_speed = [0, 0]
# simulation logging file
self.__simLogger = logging.getLogger("simulationLogger")
logLevel = logging.INFO
if debug:
logLevel = logging.DEBUG
self.__simLogger.setLevel(logLevel)
self.__experiment_name = experiment_name
outputDir = os.path.join(cl.OUTPUT_PATH, self.__experiment_name)
mkdir_p(outputDir)
# self.__nextSimFilename = getNextIDPath(SIM_LOG_PATH) + '_' + time.strftime("%Y-%m-%d_%H-%M-%S")
# simHandler = logging.FileHandler(os.path.join(SIM_LOG_PATH, self.__nextSimFilename + '_sim_debug.log'))
self.__simulationLogFile = os.path.join(outputDir, self.__experiment_name + "_sim_debug.log")
self.__simulationOutputFile = os.path.join(outputDir, self.__experiment_name + "_out.txt")
self.__simulationWeightOutputFile = os.path.join(outputDir, self.__experiment_name + "_weight_out.txt")
self.__simulationTempFile = os.path.join(outputDir, self.__experiment_name + "_temp.txt")
simHandler = logging.FileHandler(self.__simulationLogFile)
simHandler.setFormatter(cl.FORMATTER)
self.__simLogger.addHandler(simHandler)
self.__threadCamera = cameraVision(False, self.__simLogger)
self.__inbox = Inbox(self.__simLogger)
for bot in config.bots:
address = bot["address"]
self.__msgSenders[address] = MessageSender(address, bot["port"], self.__simLogger)
self.__msgReceivers[address] = MessageReceiver(address, self.__inbox, self.__simLogger)
self.__connListener = ConnectionsListener(self.__address, self.__port, self.__msgReceivers, self.__simLogger)
class Simulation(threading.Thread):
def __init__(self, thymioController, debug, experiment_name):
threading.Thread.__init__(self)
config = ConfigParser(cl.CONFIG_PATH)
self.__address = config.address
self.__port = config.port
self.__thymioController = thymioController
self.__tcPerformedAction = threading.Condition()
self.__tcPA = False
self.__msgSenders = dict()
self.__msgReceivers = dict()
self.__stopSockets = list()
self.__stopped = False
self.__previous_motor_speed = [0, 0]
# simulation logging file
self.__simLogger = logging.getLogger('simulationLogger')
logLevel = logging.INFO
if debug:
logLevel = logging.DEBUG
self.__simLogger.setLevel(logLevel)
self.__experiment_name = experiment_name
outputDir = os.path.join(cl.OUTPUT_PATH, self.__experiment_name)
mkdir_p(outputDir)
# self.__nextSimFilename = getNextIDPath(SIM_LOG_PATH) + '_' + time.strftime("%Y-%m-%d_%H-%M-%S")
# simHandler = logging.FileHandler(os.path.join(SIM_LOG_PATH, self.__nextSimFilename + '_sim_debug.log'))
self.__simulationLogFile = os.path.join(outputDir, self.__experiment_name + '_sim_debug.log')
self.__simulationOutputFile = os.path.join(outputDir, self.__experiment_name + '_out.txt')
self.__simulationWeightOutputFile = os.path.join(outputDir, self.__experiment_name + '_weight_out.txt')
self.__simulationTempFile = os.path.join(outputDir, self.__experiment_name + '_temp.txt')
simHandler = logging.FileHandler(self.__simulationLogFile)
simHandler.setFormatter(cl.FORMATTER)
self.__simLogger.addHandler(simHandler)
self.__threadCamera = cameraVision(False, self.__simLogger)
self.__inbox = Inbox(self.__simLogger)
for bot in config.bots:
address = bot["address"]
self.__msgSenders[address] = MessageSender(address, bot["port"], self.__simLogger)
self.__msgReceivers[address] = MessageReceiver(address, self.__inbox, self.__simLogger)
self.__connListener = ConnectionsListener(self.__address, self.__port, self.__msgReceivers, self.__simLogger)
def getLogger(self):
return self.__simLogger
def thymioControllerPerformedAction(self):
with self.__tcPerformedAction:
self.__tcPA = True
self.__tcPerformedAction.notify()
def __waitForControllerResponse(self):
# Wait for ThymioController response
with self.__tcPerformedAction:
while not self.__tcPA:
self.__tcPerformedAction.wait()
self.__tcPA = False
def __sendFiles(self, filepathOut, filepathLog, filepathTemp, filepathWeight):
try:
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
s.connect((cl.PC_FIXED_IP, cl.OUTPUT_FILE_RECEIVER_PORT))
# Send output file
self.__simLogger.debug("Sending file " + str(filepathOut))
sendOneMessage(s, self.__experiment_name)
fO = open(filepathOut, 'rb')
l = fO.read(1024)
while (l):
sendOneMessage(s, l)
l = fO.read(1024)
self.__simLogger.debug("End of output file")
sendOneMessage(s, cl.EOF_REACHED)
fO.close()
# Send log file
self.__simLogger.debug("Sending file " + str(filepathLog))
fL = open(filepathLog, 'rb')
l = fL.read(1024)
while (l):
sendOneMessage(s, l)
l = fL.read(1024)
self.__simLogger.debug("End of output file")
sendOneMessage(s, cl.EOF_REACHED)
fL.close()
# Send temp file
self.__simLogger.debug("Sending file " + str(filepathTemp))
fT = open(filepathTemp, 'rb')
l = fT.read(1024)
while (l):
sendOneMessage(s, l)
l = fT.read(1024)
self.__simLogger.debug("End of output file")
sendOneMessage(s, cl.EOF_REACHED)
fT.close()
# Send weight file
self.__simLogger.debug("Sending file " + str(filepathWeight))
fW = open(filepathWeight, 'rb')
l = fW.read(1024)
while (l):
sendOneMessage(s, l)
l = fW.read(1024)
fW.close()
s.shutdown(socket.SHUT_WR)
s.close()
except:
self.__simLogger.critical(
'Error while sending file: ' + str(sys.exc_info()[0]) + ' - ' + traceback.format_exc())
def __mutateMemome(self, solution, weights):
for i in range(weights):
# mutate weight adding gaussrand() * sigma
solution.memome[i] += gaussrand() * solution.sigma
if solution.memome[i] > pr.range_weights:
solution.memome[i] = pr.range_weights
elif solution.memome[i] < -pr.range_weights:
solution.memome[i] = -pr.range_weights
def __runAndEvaluate(self, evaluee, change_tau):
# Recovery period (tau timesteps)
self.__simLogger.info("Recovery period")
tau = pr.tau
if change_tau:
tau = pr.tau_goal
for i in range(0, tau):
self.__runAndEvaluateForOneTimeStep(evaluee)
# self.__simLogger.info("Timestep " + str(i) + " done")
# Evaluate (evaltime timesteps)
self.__simLogger.info("Evaluation")
fitness = [0, 0, 0, 0, 0, False]
# fitness[0] = total obstacle avoidance
# fitness[1] = total fitness looking for box
# fitness[2] = total fitness pushing the box
# fitness[3] = total fitness looking for goal
# fitness[4] = total fitness bonus
# fitness[5] = total found goal
for i in range(0, pr.eval_time):
fitness = self.__fitnessFunction(evaluee, fitness)
print("single ff -> {}".format(fitness))
# Log single fitness
self.__simLogger.info("Single fitness -> {}".format(fitness))
# Different Behaviour Different fitness
if pr.controller == 2 and pr.behaviour == 0: # Obstacle Avoidance
total_fitness = fitness[0]
elif pr.controller == 2 and pr.behaviour == 1: # Obs + Pushing Behaviour
total_fitness = sum(fitness[0:3])
else: # Foraging Task
total_fitness = sum(fitness[0:5])
print(str(total_fitness) + "\n ")
# Log total fitness
self.__simLogger.info("Fitness ->" + str(total_fitness))
return total_fitness, fitness[5]
# fitness function
def __fitnessFunction(self, evaluee, fitness):
# result_fitness[0] = obstacle avoidance / external fitness
# result_fitness[1] = fitness looking for box
# result_fitness[2] = fitness pushing the box
# result_fitness[3] = fitness looking for goal
# result_fitness[4] = found goal
result_fitness = self.__runAndEvaluateForOneTimeStep(evaluee)
fitness[0] += result_fitness[0] # total fitness obstacle avoidance
fitness[1] += result_fitness[1] # total fitness looking for box
fitness[2] += result_fitness[2] * 2 # total fitness pushing the box
fitness[3] += result_fitness[3] # total fitness looking for goal
fitness[4] += result_fitness[2] * result_fitness[3] # total fitness bonus
fitness[5] = fitness[5] or result_fitness[4] # total found goal
return fitness
def __runAndEvaluateForOneTimeStep(self, evaluee):
# Read sensors: request to ThymioController
self.__thymioController.readSensorsRequest()
self.__waitForControllerResponse()
psValues = self.__thymioController.getPSValues()
# I am not using all the sensors. I am using only NB_DIST_SENS sensors
psValues = [psValues[0], psValues[2], psValues[4], psValues[5], psValues[6]]
# return presence value from camera
presence_box = self.__threadCamera.readPuckPresence()
presence_goal = self.__threadCamera.readGoalPresence()
totalPresence = presence_box + presence_goal
for i in range(len(totalPresence)):
threshold = 1500 if i == 3 else 2000 # for bottom part better higher threshold
if totalPresence[i] > threshold:
totalPresence[i] = 1
else:
totalPresence[i] = 0
motorspeed = [0, 0]
# Neural networks
if pr.hidden_layer == 1:
# Version with one hidden layer with classes.HIDDEN_NEURONS hidden neurons
hidden_layer = [0 for x in range(cl.HIDDEN_NEURONS)]
for y in range(cl.HIDDEN_NEURONS):
for i in range(0, cl.NB_DIST_SENS): # Calculate weight only for normal sensor
normalizedSensor = min((psValues[i] - (cl.SENSOR_MAX[i] / 2)) / (cl.SENSOR_MAX[i] / 2), 1.0)
hidden_layer[y] += normalizedSensor * evaluee.memome[
i + ((cl.NN_WEIGHTS / cl.HIDDEN_NEURONS) * y)]
for i in range(0, cl.NB_CAM_SENS): # Calculate weight only for normal sensor
# normalizedSensor = min((totalPresence[i] - (cl.CAMERA_MAX[i] / 2)) / (cl.CAMERA_MAX[i] / 2),
# 1.0)
hidden_layer[y] += totalPresence[i] * evaluee.memome[
i + cl.NB_DIST_SENS + ((cl.NN_WEIGHTS / cl.HIDDEN_NEURONS) * y)]
# Adding bias weight
hidden_layer[y] += evaluee.memome[((cl.NN_WEIGHTS / cl.HIDDEN_NEURONS) * (y + 1)) - 1]
# Apply hyberbolic tangent activation function -> left and right in [-1, 1]
hidden_layer[y] = math.tanh(hidden_layer[y])
left = 0
right = 0
for i in range(0, cl.HIDDEN_NEURONS): # Calculate weight for hidden neurons
left += hidden_layer[i] * evaluee.memome[i + cl.NN_WEIGHTS]
right += hidden_layer[i] * evaluee.memome[i + cl.NN_WEIGHTS + (cl.NN_WEIGHTS_HIDDEN / 2)]
# Add bias weights
left += evaluee.memome[cl.NN_WEIGHTS + (cl.NN_WEIGHTS_HIDDEN / 2) - 1]
right += evaluee.memome[cl.TOTAL_WEIGHTS - 1]
else:
# Version without hidden layer
left = 0
right = 0
for i in range(0, cl.NB_DIST_SENS): # Calculate weight only for normal sensor
# NormalizedSensor in [-1,1]
normalizedSensor = min((psValues[i] - (cl.SENSOR_MAX[i] / 2)) / (cl.SENSOR_MAX[i] / 2), 1.0)
left += totalPresence[i] * evaluee.memome[i]
right += totalPresence[i] * evaluee.memome[i + (cl.NN_WEIGHTS_NO_HIDDEN / 2)]
for i in range(0, cl.NB_CAM_SENS): # Calculate weight only for camera sensor
# NormalizedSensor in [-1,1]
# normalizedSensor = min((totalPresence[i] - (cl.CAMERA_MAX[i] / 2)) / (cl.CAMERA_MAX[i] / 2),
# 1.0)
left += normalizedSensor * evaluee.memome[i + cl.NB_DIST_SENS]
right += normalizedSensor * evaluee.memome[i + cl.NB_DIST_SENS + (cl.NN_WEIGHTS_NO_HIDDEN / 2)]
# Add bias weights
left += evaluee.memome[(cl.NN_WEIGHTS_NO_HIDDEN / 2) - 1]
right += evaluee.memome[cl.NN_WEIGHTS_NO_HIDDEN - 1]
# Apply hyberbolic tangent activation function -> left and right in [-1, 1]
left = math.tanh(left)
right = math.tanh(right)
if left > 1 or left < -1 or right > 1 or right < -1:
self.__simLogger.warning("WUT? left = " + str(left) + ", right = " + str(right))
# Set motorspeed
motorspeed[LEFT] = int(left * pr.real_max_speed)
motorspeed[RIGHT] = int(right * pr.real_max_speed)
if (motorspeed[LEFT] != self.__previous_motor_speed[LEFT]) or (
motorspeed[RIGHT] != self.__previous_motor_speed[RIGHT]):
# Set motor speed: request to ThymioController only if the values are different from previous one
self.__thymioController.writeMotorspeedRequest(motorspeed)
self.__waitForControllerResponse()
# self.__simLogger.info("Simulation - Set motorspeed " + str(motorspeed)[1:-1])
# remember previous motor speed
self.__previous_motor_speed = motorspeed
# FITNESS FUNCTION SECTION -------------------------------------------------------------------------------------
# Calculate normalized distance to the nearest object
sensorpenalty = 0
for i in range(0, cl.NB_DIST_SENS):
if sensorpenalty < (psValues[i] / float(cl.SENSOR_MAX[i])):
sensorpenalty = (psValues[i] / float(cl.SENSOR_MAX[i]))
if sensorpenalty > 1:
sensorpenalty = 1
# Normalize all the part of the fitness from 0 to 1
normalizedSensor = (abs(motorspeed[LEFT]) + abs(motorspeed[RIGHT])) / (pr.real_max_speed * 2)
fitness_obs = float(normalizedSensor) * (1 - sensorpenalty)
found = False
# total presence is still 0 or 1
fitness_looking_for_box = totalPresence[1]
# normalized_fitness_box_pushing = 1
normalized_fitness_box_pushing = totalPresence[3]
# normalize fitness_looking_for_goal 0 -> 1 (minimum is 0 so i can not write it )
normalized_fitness_looking_goal = totalPresence[5]
total_area_goal = sum(presence_goal[0:3])
# Reach the goal
if total_area_goal > 18500 and normalized_fitness_box_pushing == 1:
found = True
# Make sound
self.__thymioController.soundRequest(
[cl.SOUND]) # system sound -> value from 0 to 7 (4 = free-fall (scary) sound)
self.__waitForControllerResponse()
# print goal
self.__simLogger.info("GOAL REACHED\t")
fitness_result = (
fitness_obs, fitness_looking_for_box, normalized_fitness_box_pushing,
normalized_fitness_looking_goal, found)
return fitness_result
def run(self):
# Start ConnectionsListener
self.__connListener.start()
# Set connections for stopping the receivers
for i in range(0, len(self.__msgReceivers)):
stopSocket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
stopSocket.connect((cl.LOCALHOST, self.__port))
self.__stopSockets.append(stopSocket)
# Start message receivers
for addr in self.__msgReceivers:
self.__msgReceivers[addr].start()
# Start message senders
for addr in self.__msgSenders:
self.__msgSenders[addr].start()
# Wait for the simulation to be set on the controller
self.__waitForControllerResponse()
# Set color: request to ThymioController
self.__thymioController.writeColorRequest([0, 0, 0, 0]) # Switch off all the leds
self.__waitForControllerResponse()
# Starting CPU timer -> for temperature
t_start = time.clock()
# Start camera thread
self.__threadCamera.start()
# wait until camera starts
time.sleep(5)
# set correct weights
if pr.hidden_layer == 1:
weights = cl.TOTAL_WEIGHTS
else:
weights = cl.NN_WEIGHTS_NO_HIDDEN
# Set parameters
lifetime = pr.max_robot_lifetime if pr.evolution == 1 else pr.total_evals # when no evolution, robot life
# is whole experiment length
collected_memomes = [cl.RobotMemomeDataMessage(0.0, [0.0 for i in range(weights)]) for i in
range(pr.collected_memomes_max)]
evals = 0 # current evaluation
generation = 0 # current generation
champion = cl.Candidate([0.0 for i in range(weights)], 0.0, 0.0)
actual_weights = cl.Candidate([0.0 for i in range(weights)], 0.0, 0.0)
total_goal = 0 # count numbers of goals
# start main loop
while evals < pr.total_evals:
generation += 1
self.__simLogger.info("########## GENERATION " + str(generation) + " ##########")
self.__simLogger.info("Current champion: \nMemome: " + str(champion.memome) + "\nSigma: " +
str(champion.sigma) + "\nFitness: " + str(champion.fitness))
pr.collected_memomes_total = 0
# INIT NEW MEMOME
# if controller == 3 run experiment with fixed weight
if pr.controller == 3:
champion.memome = pr.fixed_weight
else:
# Normal experiment with random initialize
tmp_weight = [random.uniform(-1 * pr.range_weights, pr.range_weights) for x in range(weights)]
# Set random neural network weights
for i in range(weights):
champion.memome[i] = tmp_weight[i]
champion.sigma = pr.sigmainitial
try:
change_tau = False # Track if i found the goal
# Evaluate the new individual
champion.fitness, goalI = self.__runAndEvaluate(copy.deepcopy(champion), change_tau)
self.__simLogger.info("New champion: \nMemome: " + str(champion.memome) + "\nSigma: " +
str(champion.sigma) + "\nFitness: " + str(champion.fitness))
# DURING INDIVIDUAL LIFE
for l in range(lifetime):
self.__simLogger.info("@@@@@ EVALUATION " + str(evals) + " @@@@@")
evals += 1
goal = False
# Choose one between: Evolving controller | Random controller | Expert controller | Fixed controller
if pr.controller == 2: # Evolving controller
# Choose one between: Reevaluation | Social learning | Individual learning
if random.random() <= 0.2:
# Reevaluation
self.__simLogger.info("----- REEVALUATION -----")
self.__simLogger.info("Old fitness = " + str(champion.fitness))
fitness, goal = self.__runAndEvaluate(champion, change_tau)
self.__simLogger.info("Reevaluated fitness = " + str(fitness))
champion.fitness = champion.fitness * pr.re_weight + fitness * (1 - pr.re_weight)
self.__simLogger.info("New fitness = " + str(champion.fitness))
# Save neural network weights
actual_weights = champion
else:
if pr.collected_memomes_total > 0 and pr.sociallearning == 1 and random.random() <= 0.3:
# Social learning
self.__simLogger.info("----- SOCIAL LEARNING -----")
socialChallenger = copy.deepcopy(
champion) # Deep copy: we don't want to change the champion
# Memome Crossover with last memotype in collected_memomes (LIFO order)
lastCollectedMemome = collected_memomes[pr.collected_memomes_total - 1].memome
# 25% probability (One-Point crossover) average two memome
if random.random() <= 0.75:
for i in range(weights):
# overwrite value on champion
socialChallenger.memome[i] = lastCollectedMemome[i]
else:
# One-point crossover from 0 to cutting_value
# cutting_value = random.randrange(0, cl.NMBRWEIGHTS)
# for i in range(cl.NMBRWEIGHTS):
# if (socialChallenger.memome[i] != 100.0 and lastCollectedMemome[i] != 100.0):
# # If sensor is enabled on both champion and lastCollectedMemome -> overwrite
# if i <= cutting_value:
# # value on champion
# socialChallenger.memome[i] = lastCollectedMemome[i]
# Average two memome
for i in range(weights):
socialChallenger.memome[i] = (lastCollectedMemome[i] + socialChallenger.memome[
i]) / 2
pr.collected_memomes_total -= 1
socialChallenger.fitness, goal = self.__runAndEvaluate(socialChallenger, change_tau)
self.__simLogger.info("Social challenger memome = " + str(socialChallenger.memome))
self.__simLogger.info("Social challenger fitness = " + str(socialChallenger.fitness))
# Save neural network weights
actual_weights = socialChallenger
if socialChallenger.fitness > champion.fitness:
self.__simLogger.info("Social challenger IS better -> Becomes champion")
champion = socialChallenger
champion.sigma = pr.sigma_min
else:
self.__simLogger.info("Social challenger NOT better -> Sigma is doubled")
champion.sigma = champion.sigma * pr.sigma_increase
if champion.sigma > pr.sigma_max:
champion.sigma = pr.sigma_max
self.__simLogger.info("New sigma = " + str(champion.sigma))
else:
# Individual learning
self.__simLogger.info("----- INDIVIDUAL LEARNING -----")
if pr.lifetimelearning == 1:
challenger = copy.deepcopy(
champion) # Deep copy: we don't want to mutate the champion
self.__simLogger.info("Sigma = " + str(champion.sigma))
self.__simLogger.info(
"Challenger memome before mutation = " + str(challenger.memome))
self.__mutateMemome(challenger, weights)
self.__simLogger.info("Mutated challenger memome = " + str(challenger.memome))
challenger.fitness, goal = self.__runAndEvaluate(challenger, change_tau)
self.__simLogger.info("Mutated challenger fitness = " + str(challenger.fitness) +
" VS Champion fitness = " + str(champion.fitness))
# Save neural network weights
actual_weights = challenger
if challenger.fitness > champion.fitness:
self.__simLogger.info("Challenger IS better -> Becomes champion")
champion = challenger
champion.sigma = pr.sigma_min
else:
self.__simLogger.info("Challenger NOT better -> Sigma is doubled")
champion.sigma = champion.sigma * pr.sigma_increase
if champion.sigma > pr.sigma_max: # boundary rule
champion.sigma = pr.sigma_max
self.__simLogger.info("New sigma = " + str(champion.sigma))
elif pr.controller == 0: # Random controller
self.__simLogger.info("----- RANDOM CONTROLLER -----")
fitness, goal = self.__runAndEvaluate(champion, change_tau)
self.__simLogger.info("Random Controller fitness = " + str(fitness))
# Save neural network weights
actual_weights = champion
# Random Memome
tmp_weight = [random.uniform(-1 * pr.range_weights, pr.range_weights) for x in range(weights)]
# Set random neural network weights
for i in range(weights):
champion.memome[i] = tmp_weight[i]
elif pr.controller == 3: # Fixed controller
self.__simLogger.info("----- FIXED CONTROLLER -----")
fitness, goal = self.__runAndEvaluate(champion, change_tau)
self.__simLogger.info("Fixed Controller fitness = " + str(fitness))
elif pr.controller == 1: # Expert coontroller
self.__simLogger.info("----- EXPERT CONTROLLER -----")
# TO DO
self.__simLogger.info("Current champion: \nMemome: " + str(champion.memome) + "\nSigma: " +
str(champion.sigma) + "\nFitness: " + str(champion.fitness))
# Make sound if I found the goal
if goal or goalI:
change_tau = True # Longer recovery time
total_goal += 1 # Increase numbers of goals
else:
change_tau = False # Make sure that tau is correct if I did't find the goal
# Write output: open file to append the values
with open(self.__simulationOutputFile, 'a') as outputFile:
outputFile.write(str(evals) + " \t " + str(generation) + " \t " + str(l) + " \t " + str(
champion.fitness))
if goal or goalI:
outputFile.write("\tGOAL n: " + str(total_goal))
outputFile.write("\n")
# Write weight output: open file to append the values
with open(self.__simulationWeightOutputFile, 'a') as outputFile:
outputFile.write(str(evals) + " \t " + str(generation) + " \t " + str(l) + " \t " + str(
actual_weights.fitness) + "\nMemome: " + str(actual_weights.memome) +
"\nChampion: " + str(champion.memome))
outputFile.write("\n")
# Retrieve temperature value
# Read sensors: request to ThymioController
self.__thymioController.readTemperatureRequest()
self.__waitForControllerResponse()
temperature = self.__thymioController.getTemperature()
# second from start
t_from_start = time.clock() - t_start
# Write temp output: open file to append the values
with open(self.__simulationTempFile, 'a') as outputFile:
outputFile.write(str(evals) + " \t Temperature -> " + str(temperature[0]) + " \t after " + str(
t_from_start) + " \t seconds")
outputFile.write("\n")
# Send messages: tell MessageSenders to do that
self.__simLogger.info("Broadcasting messages...")
# Different Behaviour Different max_fitness
if pr.behaviour == 0:
real_max_fitness = pr.obs_max_fitness
elif pr.behaviour == 1:
real_max_fitness = pr.push_max_fitness
else:
real_max_fitness = pr.max_fitness
if pr.sociallearning == 1 and (champion.fitness / real_max_fitness) > pr.threshold:
messageMem = cl.RobotMemomeDataMessage(champion.fitness, champion.memome)
self.__simLogger.info("Broadcast memome = " + str(messageMem.memome))
for addr in self.__msgSenders:
self.__msgSenders[addr].outboxAppend(messageMem)
# Read received messages from Inbox
receivedMsgs = self.__inbox.popAll()
self.__simLogger.info("Reading " + str(len(receivedMsgs)) + " received messages")
for rmsg in receivedMsgs:
if type(rmsg) is cl.RobotMemomeDataMessage:
if pr.collected_memomes_total < pr.collected_memomes_max:
self.__simLogger.info("Received memome = " + str(rmsg.memome))
collected_memomes[pr.collected_memomes_total] = rmsg
pr.collected_memomes_total += 1
else:
self.__simLogger.warning("WUT? Received stuff = " + str(rmsg))
# check if camera thread is still alive
# Camera could raise one exception (corrupted image). If this happens stop that thread and start
# one new
if not self.__threadCamera.isAlive():
self.__threadCamera.stop()
self.__threadCamera.join()
self.__threadCamera = cameraVision(False, self.__simLogger)
self.__threadCamera.start()
self.__simLogger.warning("Reanimating Camera Thread")
except Exception as e:
self.__simLogger.critical("Some exception: " + str(e) + str(
sys.exc_info()[0]) + ' - ' + traceback.format_exc())
self.__thymioController.stopThymioRequest()
break
self.__simLogger.info("End of while loop: " + str(evals) + " >= " + str(pr.total_evals))
self.stop()
self.__simLogger.info("_____END OF SIMULATION_____\n")
def isStopped(self):
return self.__stopped
def stop(self):
if self.__stopped:
self.__simLogger.info('Simulation already stopped.')
return
self.__stopped = True
# Send log files to server.
self.__simLogger.info("Sending Files...\n")
self.__sendFiles(self.__simulationOutputFile, self.__simulationLogFile, self.__simulationTempFile,
self.__simulationWeightOutputFile)
# Stop Thymio from moving
self.__thymioController.stopThymioRequest()
# Stopping all the message senders: no more outgoing messages
for addr in self.__msgSenders:
self.__simLogger.info('Killing Sender ' + addr)
self.__msgSenders[addr].stop()
self.__msgSenders[addr].join()
self.__simLogger.info('All MessageSenders: KILLED')
# Stopping connections listener: no more incoming connections
self.__connListener.stop()
self.__connListener.join()
self.__simLogger.info('ConnectionsListener: KILLED')
# Stopping camera thread
self.__threadCamera.stop()
self.__threadCamera.join()
self.__simLogger.info('CameraThread: KILLED')
# Stopping all the message receivers: no more incoming messages
i = 0
for addr in self.__msgReceivers:
self.__simLogger.info('Killing Receiver ' + addr)
self.__msgReceivers[addr].stop()
sendOneMessage(self.__stopSockets[i], 'STOP') # Send stop messages
self.__msgReceivers[addr].join()
self.__stopSockets[i].close()
i += 1
self.__simLogger.info('All MessageReceivers: KILLED')
class Simulation(threading.Thread):
def __init__(self, thymioController, debug, experiment_name):
threading.Thread.__init__(self)
config = ConfigParser(cl.CONFIG_PATH)
self.__address = config.address
self.__port = config.port
self.__thymioController = thymioController
self.__tcPerformedAction = threading.Condition()
self.__tcPA = False
self.__msgSenders = dict()
self.__msgReceivers = dict()
self.__stopSockets = list()
self.__stopped = False
self.__previous_motor_speed = [0, 0]
# simulation logging file
self.__simLogger = logging.getLogger("simulationLogger")
logLevel = logging.INFO
if debug:
logLevel = logging.DEBUG
self.__simLogger.setLevel(logLevel)
self.__experiment_name = experiment_name
outputDir = os.path.join(cl.OUTPUT_PATH, self.__experiment_name)
mkdir_p(outputDir)
# self.__nextSimFilename = getNextIDPath(SIM_LOG_PATH) + '_' + time.strftime("%Y-%m-%d_%H-%M-%S")
# simHandler = logging.FileHandler(os.path.join(SIM_LOG_PATH, self.__nextSimFilename + '_sim_debug.log'))
self.__simulationLogFile = os.path.join(outputDir, self.__experiment_name + "_sim_debug.log")
self.__simulationOutputFile = os.path.join(outputDir, self.__experiment_name + "_out.txt")
self.__simulationWeightOutputFile = os.path.join(outputDir, self.__experiment_name + "_weight_out.txt")
self.__simulationTempFile = os.path.join(outputDir, self.__experiment_name + "_temp.txt")
simHandler = logging.FileHandler(self.__simulationLogFile)
simHandler.setFormatter(cl.FORMATTER)
self.__simLogger.addHandler(simHandler)
self.__threadCamera = cameraVision(False, self.__simLogger)
self.__inbox = Inbox(self.__simLogger)
for bot in config.bots:
address = bot["address"]
self.__msgSenders[address] = MessageSender(address, bot["port"], self.__simLogger)
self.__msgReceivers[address] = MessageReceiver(address, self.__inbox, self.__simLogger)
self.__connListener = ConnectionsListener(self.__address, self.__port, self.__msgReceivers, self.__simLogger)
def getLogger(self):
return self.__simLogger
def thymioControllerPerformedAction(self):
with self.__tcPerformedAction:
self.__tcPA = True
self.__tcPerformedAction.notify()
def __waitForControllerResponse(self):
# Wait for ThymioController response
with self.__tcPerformedAction:
while not self.__tcPA:
self.__tcPerformedAction.wait()
self.__tcPA = False
def __sendFiles(self, filepathOut, filepathLog, filepathTemp, filepathWeight):
try:
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
s.connect((cl.PC_FIXED_IP, cl.OUTPUT_FILE_RECEIVER_PORT))
# Send output file
self.__simLogger.debug("Sending file " + str(filepathOut))
sendOneMessage(s, self.__experiment_name)
fO = open(filepathOut, "rb")
l = fO.read(1024)
while l:
sendOneMessage(s, l)
l = fO.read(1024)
self.__simLogger.debug("End of output file")
sendOneMessage(s, cl.EOF_REACHED)
fO.close()
# Send log file
self.__simLogger.debug("Sending file " + str(filepathLog))
fL = open(filepathLog, "rb")
l = fL.read(1024)
while l:
sendOneMessage(s, l)
l = fL.read(1024)
self.__simLogger.debug("End of output file")
sendOneMessage(s, cl.EOF_REACHED)
fL.close()
# Send temp file
self.__simLogger.debug("Sending file " + str(filepathTemp))
fT = open(filepathTemp, "rb")
l = fT.read(1024)
while l:
sendOneMessage(s, l)
l = fT.read(1024)
self.__simLogger.debug("End of output file")
sendOneMessage(s, cl.EOF_REACHED)
fT.close()
# Send weight file
self.__simLogger.debug("Sending file " + str(filepathWeight))
fW = open(filepathWeight, "rb")
l = fW.read(1024)
while l:
sendOneMessage(s, l)
l = fW.read(1024)
fW.close()
s.shutdown(socket.SHUT_WR)
s.close()
except:
self.__simLogger.critical(
"Error while sending file: " + str(sys.exc_info()[0]) + " - " + traceback.format_exc()
)
def __mutateMemome(self, solution, weights):
for i in range(weights):
# mutate weight adding gaussrand() * sigma
solution.memome[i] += gaussrand() * solution.sigma
if solution.memome[i] > pr.range_weights:
solution.memome[i] = pr.range_weights
elif solution.memome[i] < -pr.range_weights:
solution.memome[i] = -pr.range_weights
def __runAndEvaluate(self, evaluee, change_tau):
# Recovery period (tau timesteps)
self.__simLogger.info("Recovery period")
tau = pr.tau
if change_tau:
tau = pr.tau_goal
for i in range(0, tau):
self.__runAndEvaluateForOneTimeStep(evaluee)
# self.__simLogger.info("Timestep " + str(i) + " done")
# Evaluate (evaltime timesteps)
self.__simLogger.info("Evaluation")
fitness = [0, 0, 0, 0, 0, False]
# fitness[0] = total obstacle avoidance
# fitness[1] = total fitness looking for box
# fitness[2] = total fitness pushing the box
# fitness[3] = total fitness looking for goal
# fitness[4] = total fitness bonus
# fitness[5] = total found goal
for i in range(0, pr.eval_time):
fitness = self.__fitnessFunction(evaluee, fitness)
print("single ff -> {}".format(fitness))
# Log single fitness
self.__simLogger.info("Single fitness -> {}".format(fitness))
# Different Behaviour Different fitness
if pr.controller == 2 and pr.behaviour == 0: # Obstacle Avoidance
total_fitness = fitness[0]
elif pr.controller == 2 and pr.behaviour == 1: # Obs + Pushing Behaviour
total_fitness = sum(fitness[0:3])
else: # Foraging Task
total_fitness = sum(fitness[0:5])
print(str(total_fitness) + "\n ")
# Log total fitness
self.__simLogger.info("Fitness ->" + str(total_fitness))
return total_fitness, fitness[5]
# fitness function
def __fitnessFunction(self, evaluee, fitness):
# result_fitness[0] = obstacle avoidance / external fitness
# result_fitness[1] = fitness looking for box
# result_fitness[2] = fitness pushing the box
# result_fitness[3] = fitness looking for goal
# result_fitness[4] = found goal
result_fitness = self.__runAndEvaluateForOneTimeStep(evaluee)
fitness[0] += result_fitness[0] # total fitness obstacle avoidance
fitness[1] += result_fitness[1] # total fitness looking for box
fitness[2] += result_fitness[2] * 2 # total fitness pushing the box
fitness[3] += result_fitness[3] # total fitness looking for goal
fitness[4] += result_fitness[2] * result_fitness[3] # total fitness bonus
fitness[5] = fitness[5] or result_fitness[4] # total found goal
return fitness
def __runAndEvaluateForOneTimeStep(self, evaluee):
# Read sensors: request to ThymioController
self.__thymioController.readSensorsRequest()
self.__waitForControllerResponse()
psValues = self.__thymioController.getPSValues()
# I am not using all the sensors. I am using only NB_DIST_SENS sensors
psValues = [psValues[0], psValues[2], psValues[4], psValues[5], psValues[6]]
# return presence value from camera
presence_box = self.__threadCamera.readPuckPresence()
presence_goal = self.__threadCamera.readGoalPresence()
totalPresence = presence_box + presence_goal
for i in range(len(totalPresence)):
threshold = 1500 if i == 3 else 2000 # for bottom part better higher threshold
if totalPresence[i] > threshold:
totalPresence[i] = 1
else:
totalPresence[i] = 0
motorspeed = [0, 0]
# Neural networks
if pr.hidden_layer == 1:
# Version with one hidden layer with classes.HIDDEN_NEURONS hidden neurons
hidden_layer = [0 for x in range(cl.HIDDEN_NEURONS)]
for y in range(cl.HIDDEN_NEURONS):
for i in range(0, cl.NB_DIST_SENS): # Calculate weight only for normal sensor
normalizedSensor = min((psValues[i] - (cl.SENSOR_MAX[i] / 2)) / (cl.SENSOR_MAX[i] / 2), 1.0)
hidden_layer[y] += normalizedSensor * evaluee.memome[i + ((cl.NN_WEIGHTS / cl.HIDDEN_NEURONS) * y)]
for i in range(0, cl.NB_CAM_SENS): # Calculate weight only for normal sensor
# normalizedSensor = min((totalPresence[i] - (cl.CAMERA_MAX[i] / 2)) / (cl.CAMERA_MAX[i] / 2),
# 1.0)
hidden_layer[y] += (
totalPresence[i]
* evaluee.memome[i + cl.NB_DIST_SENS + ((cl.NN_WEIGHTS / cl.HIDDEN_NEURONS) * y)]
)
# Adding bias weight
hidden_layer[y] += evaluee.memome[((cl.NN_WEIGHTS / cl.HIDDEN_NEURONS) * (y + 1)) - 1]
# Apply hyberbolic tangent activation function -> left and right in [-1, 1]
hidden_layer[y] = math.tanh(hidden_layer[y])
left = 0
right = 0
for i in range(0, cl.HIDDEN_NEURONS): # Calculate weight for hidden neurons
left += hidden_layer[i] * evaluee.memome[i + cl.NN_WEIGHTS]
right += hidden_layer[i] * evaluee.memome[i + cl.NN_WEIGHTS + (cl.NN_WEIGHTS_HIDDEN / 2)]
# Add bias weights
left += evaluee.memome[cl.NN_WEIGHTS + (cl.NN_WEIGHTS_HIDDEN / 2) - 1]
right += evaluee.memome[cl.TOTAL_WEIGHTS - 1]
else:
# Version without hidden layer
left = 0
right = 0
for i in range(0, cl.NB_DIST_SENS): # Calculate weight only for normal sensor
# NormalizedSensor in [-1,1]
normalizedSensor = min((psValues[i] - (cl.SENSOR_MAX[i] / 2)) / (cl.SENSOR_MAX[i] / 2), 1.0)
left += totalPresence[i] * evaluee.memome[i]
right += totalPresence[i] * evaluee.memome[i + (cl.NN_WEIGHTS_NO_HIDDEN / 2)]
for i in range(0, cl.NB_CAM_SENS): # Calculate weight only for camera sensor
# NormalizedSensor in [-1,1]
# normalizedSensor = min((totalPresence[i] - (cl.CAMERA_MAX[i] / 2)) / (cl.CAMERA_MAX[i] / 2),
# 1.0)
left += normalizedSensor * evaluee.memome[i + cl.NB_DIST_SENS]
right += normalizedSensor * evaluee.memome[i + cl.NB_DIST_SENS + (cl.NN_WEIGHTS_NO_HIDDEN / 2)]
# Add bias weights
left += evaluee.memome[(cl.NN_WEIGHTS_NO_HIDDEN / 2) - 1]
right += evaluee.memome[cl.NN_WEIGHTS_NO_HIDDEN - 1]
# Apply hyberbolic tangent activation function -> left and right in [-1, 1]
left = math.tanh(left)
right = math.tanh(right)
if left > 1 or left < -1 or right > 1 or right < -1:
self.__simLogger.warning("WUT? left = " + str(left) + ", right = " + str(right))
# Set motorspeed
motorspeed[LEFT] = int(left * pr.real_max_speed)
motorspeed[RIGHT] = int(right * pr.real_max_speed)
if (motorspeed[LEFT] != self.__previous_motor_speed[LEFT]) or (
motorspeed[RIGHT] != self.__previous_motor_speed[RIGHT]
):
# Set motor speed: request to ThymioController only if the values are different from previous one
self.__thymioController.writeMotorspeedRequest(motorspeed)
self.__waitForControllerResponse()
# self.__simLogger.info("Simulation - Set motorspeed " + str(motorspeed)[1:-1])
# remember previous motor speed
self.__previous_motor_speed = motorspeed
# FITNESS FUNCTION SECTION -------------------------------------------------------------------------------------
# Calculate normalized distance to the nearest object
sensorpenalty = 0
for i in range(0, cl.NB_DIST_SENS):
if sensorpenalty < (psValues[i] / float(cl.SENSOR_MAX[i])):
sensorpenalty = psValues[i] / float(cl.SENSOR_MAX[i])
if sensorpenalty > 1:
sensorpenalty = 1
# Normalize all the part of the fitness from 0 to 1
normalizedSensor = (abs(motorspeed[LEFT]) + abs(motorspeed[RIGHT])) / (pr.real_max_speed * 2)
fitness_obs = float(normalizedSensor) * (1 - sensorpenalty)
found = False
# total presence is still 0 or 1
fitness_looking_for_box = totalPresence[1]
# normalized_fitness_box_pushing = 1
normalized_fitness_box_pushing = totalPresence[3]
# normalize fitness_looking_for_goal 0 -> 1 (minimum is 0 so i can not write it )
normalized_fitness_looking_goal = totalPresence[5]
total_area_goal = sum(presence_goal[0:3])
# Reach the goal
if total_area_goal > 18500 and normalized_fitness_box_pushing == 1:
found = True
# Make sound
self.__thymioController.soundRequest(
[cl.SOUND]
) # system sound -> value from 0 to 7 (4 = free-fall (scary) sound)
self.__waitForControllerResponse()
# print goal
self.__simLogger.info("GOAL REACHED\t")
fitness_result = (
fitness_obs,
fitness_looking_for_box,
normalized_fitness_box_pushing,
normalized_fitness_looking_goal,
found,
)
return fitness_result
def run(self):
# Start ConnectionsListener
self.__connListener.start()
# Set connections for stopping the receivers
for i in range(0, len(self.__msgReceivers)):
stopSocket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
stopSocket.connect((cl.LOCALHOST, self.__port))
self.__stopSockets.append(stopSocket)
# Start message receivers
for addr in self.__msgReceivers:
self.__msgReceivers[addr].start()
# Start message senders
for addr in self.__msgSenders:
self.__msgSenders[addr].start()
# Wait for the simulation to be set on the controller
self.__waitForControllerResponse()
# Set color: request to ThymioController
self.__thymioController.writeColorRequest([0, 0, 0, 0]) # Switch off all the leds
self.__waitForControllerResponse()
# Starting CPU timer -> for temperature
t_start = time.clock()
# Start camera thread
self.__threadCamera.start()
# wait until camera starts
time.sleep(5)
# set correct weights
if pr.hidden_layer == 1:
weights = cl.TOTAL_WEIGHTS
else:
weights = cl.NN_WEIGHTS_NO_HIDDEN
# Set parameters
lifetime = pr.max_robot_lifetime if pr.evolution == 1 else pr.total_evals # when no evolution, robot life
# is whole experiment length
collected_memomes = [
cl.RobotMemomeDataMessage(0.0, [0.0 for i in range(weights)]) for i in range(pr.collected_memomes_max)
]
evals = 0 # current evaluation
generation = 0 # current generation
champion = cl.Candidate([0.0 for i in range(weights)], 0.0, 0.0)
actual_weights = cl.Candidate([0.0 for i in range(weights)], 0.0, 0.0)
total_goal = 0 # count numbers of goals
# start main loop
while evals < pr.total_evals:
generation += 1
self.__simLogger.info("########## GENERATION " + str(generation) + " ##########")
self.__simLogger.info(
"Current champion: \nMemome: "
+ str(champion.memome)
+ "\nSigma: "
+ str(champion.sigma)
+ "\nFitness: "
+ str(champion.fitness)
)
pr.collected_memomes_total = 0
# INIT NEW MEMOME
# if controller == 3 run experiment with fixed weight
if pr.controller == 3:
champion.memome = pr.fixed_weight
else:
# Normal experiment with random initialize
tmp_weight = [random.uniform(-1 * pr.range_weights, pr.range_weights) for x in range(weights)]
# Set random neural network weights
for i in range(weights):
champion.memome[i] = tmp_weight[i]
champion.sigma = pr.sigmainitial
try:
change_tau = False # Track if i found the goal
# Evaluate the new individual
champion.fitness, goalI = self.__runAndEvaluate(copy.deepcopy(champion), change_tau)
self.__simLogger.info(
"New champion: \nMemome: "
+ str(champion.memome)
+ "\nSigma: "
+ str(champion.sigma)
+ "\nFitness: "
+ str(champion.fitness)
)
# DURING INDIVIDUAL LIFE
for l in range(lifetime):
self.__simLogger.info("@@@@@ EVALUATION " + str(evals) + " @@@@@")
evals += 1
goal = False
# Choose one between: Evolving controller | Random controller | Expert controller | Fixed controller
if pr.controller == 2: # Evolving controller
# Choose one between: Reevaluation | Social learning | Individual learning
if random.random() <= 0.2:
# Reevaluation
self.__simLogger.info("----- REEVALUATION -----")
self.__simLogger.info("Old fitness = " + str(champion.fitness))
fitness, goal = self.__runAndEvaluate(champion, change_tau)
self.__simLogger.info("Reevaluated fitness = " + str(fitness))
champion.fitness = champion.fitness * pr.re_weight + fitness * (1 - pr.re_weight)
self.__simLogger.info("New fitness = " + str(champion.fitness))
# Save neural network weights
actual_weights = champion
else:
if pr.collected_memomes_total > 0 and pr.sociallearning == 1 and random.random() <= 0.3:
# Social learning
self.__simLogger.info("----- SOCIAL LEARNING -----")
socialChallenger = copy.deepcopy(
champion
) # Deep copy: we don't want to change the champion
# Memome Crossover with last memotype in collected_memomes (LIFO order)
lastCollectedMemome = collected_memomes[pr.collected_memomes_total - 1].memome
# 25% probability (One-Point crossover) average two memome
if random.random() <= 0.75:
for i in range(weights):
# overwrite value on champion
socialChallenger.memome[i] = lastCollectedMemome[i]
else:
# One-point crossover from 0 to cutting_value
# cutting_value = random.randrange(0, cl.NMBRWEIGHTS)
# for i in range(cl.NMBRWEIGHTS):
# if (socialChallenger.memome[i] != 100.0 and lastCollectedMemome[i] != 100.0):
# # If sensor is enabled on both champion and lastCollectedMemome -> overwrite
# if i <= cutting_value:
# # value on champion
# socialChallenger.memome[i] = lastCollectedMemome[i]
# Average two memome
for i in range(weights):
socialChallenger.memome[i] = (
lastCollectedMemome[i] + socialChallenger.memome[i]
) / 2
pr.collected_memomes_total -= 1
socialChallenger.fitness, goal = self.__runAndEvaluate(socialChallenger, change_tau)
self.__simLogger.info("Social challenger memome = " + str(socialChallenger.memome))
self.__simLogger.info("Social challenger fitness = " + str(socialChallenger.fitness))
# Save neural network weights
actual_weights = socialChallenger
if socialChallenger.fitness > champion.fitness:
self.__simLogger.info("Social challenger IS better -> Becomes champion")
champion = socialChallenger
champion.sigma = pr.sigma_min
else:
self.__simLogger.info("Social challenger NOT better -> Sigma is doubled")
champion.sigma = champion.sigma * pr.sigma_increase
if champion.sigma > pr.sigma_max:
champion.sigma = pr.sigma_max
self.__simLogger.info("New sigma = " + str(champion.sigma))
else:
# Individual learning
self.__simLogger.info("----- INDIVIDUAL LEARNING -----")
if pr.lifetimelearning == 1:
challenger = copy.deepcopy(
champion
) # Deep copy: we don't want to mutate the champion
self.__simLogger.info("Sigma = " + str(champion.sigma))
self.__simLogger.info(
"Challenger memome before mutation = " + str(challenger.memome)
)
self.__mutateMemome(challenger, weights)
self.__simLogger.info("Mutated challenger memome = " + str(challenger.memome))
challenger.fitness, goal = self.__runAndEvaluate(challenger, change_tau)
self.__simLogger.info(
"Mutated challenger fitness = "
+ str(challenger.fitness)
+ " VS Champion fitness = "
+ str(champion.fitness)
)
# Save neural network weights
actual_weights = challenger
if challenger.fitness > champion.fitness:
self.__simLogger.info("Challenger IS better -> Becomes champion")
champion = challenger
champion.sigma = pr.sigma_min
else:
self.__simLogger.info("Challenger NOT better -> Sigma is doubled")
champion.sigma = champion.sigma * pr.sigma_increase
if champion.sigma > pr.sigma_max: # boundary rule
champion.sigma = pr.sigma_max
self.__simLogger.info("New sigma = " + str(champion.sigma))
elif pr.controller == 0: # Random controller
self.__simLogger.info("----- RANDOM CONTROLLER -----")
fitness, goal = self.__runAndEvaluate(champion, change_tau)
self.__simLogger.info("Random Controller fitness = " + str(fitness))
# Save neural network weights
actual_weights = champion
# Random Memome
tmp_weight = [random.uniform(-1 * pr.range_weights, pr.range_weights) for x in range(weights)]
# Set random neural network weights
for i in range(weights):
champion.memome[i] = tmp_weight[i]
elif pr.controller == 3: # Fixed controller
self.__simLogger.info("----- FIXED CONTROLLER -----")
fitness, goal = self.__runAndEvaluate(champion, change_tau)
self.__simLogger.info("Fixed Controller fitness = " + str(fitness))
elif pr.controller == 1: # Expert coontroller
self.__simLogger.info("----- EXPERT CONTROLLER -----")
# TO DO
self.__simLogger.info(
"Current champion: \nMemome: "
+ str(champion.memome)
+ "\nSigma: "
+ str(champion.sigma)
+ "\nFitness: "
+ str(champion.fitness)
)
# Make sound if I found the goal
if goal or goalI:
change_tau = True # Longer recovery time
total_goal += 1 # Increase numbers of goals
else:
change_tau = False # Make sure that tau is correct if I did't find the goal
# Write output: open file to append the values
with open(self.__simulationOutputFile, "a") as outputFile:
outputFile.write(
str(evals) + " \t " + str(generation) + " \t " + str(l) + " \t " + str(champion.fitness)
)
if goal or goalI:
outputFile.write("\tGOAL n: " + str(total_goal))
outputFile.write("\n")
# Write weight output: open file to append the values
with open(self.__simulationWeightOutputFile, "a") as outputFile:
outputFile.write(
str(evals)
+ " \t "
+ str(generation)
+ " \t "
+ str(l)
+ " \t "
+ str(actual_weights.fitness)
+ "\nMemome: "
+ str(actual_weights.memome)
+ "\nChampion: "
+ str(champion.memome)
)
outputFile.write("\n")
# Retrieve temperature value
# Read sensors: request to ThymioController
self.__thymioController.readTemperatureRequest()
self.__waitForControllerResponse()
temperature = self.__thymioController.getTemperature()
# second from start
t_from_start = time.clock() - t_start
# Write temp output: open file to append the values
with open(self.__simulationTempFile, "a") as outputFile:
outputFile.write(
str(evals)
+ " \t Temperature -> "
+ str(temperature[0])
+ " \t after "
+ str(t_from_start)
+ " \t seconds"
)
outputFile.write("\n")
# Send messages: tell MessageSenders to do that
self.__simLogger.info("Broadcasting messages...")
# Different Behaviour Different max_fitness
if pr.behaviour == 0:
real_max_fitness = pr.obs_max_fitness
elif pr.behaviour == 1:
real_max_fitness = pr.push_max_fitness
else:
real_max_fitness = pr.max_fitness
if pr.sociallearning == 1 and (champion.fitness / real_max_fitness) > pr.threshold:
messageMem = cl.RobotMemomeDataMessage(champion.fitness, champion.memome)
self.__simLogger.info("Broadcast memome = " + str(messageMem.memome))
for addr in self.__msgSenders:
self.__msgSenders[addr].outboxAppend(messageMem)
# Read received messages from Inbox
receivedMsgs = self.__inbox.popAll()
self.__simLogger.info("Reading " + str(len(receivedMsgs)) + " received messages")
for rmsg in receivedMsgs:
if type(rmsg) is cl.RobotMemomeDataMessage:
if pr.collected_memomes_total < pr.collected_memomes_max:
self.__simLogger.info("Received memome = " + str(rmsg.memome))
collected_memomes[pr.collected_memomes_total] = rmsg
pr.collected_memomes_total += 1
else:
self.__simLogger.warning("WUT? Received stuff = " + str(rmsg))
# check if camera thread is still alive
# Camera could raise one exception (corrupted image). If this happens stop that thread and start
# one new
if not self.__threadCamera.isAlive():
self.__threadCamera.stop()
self.__threadCamera.join()
self.__threadCamera = cameraVision(False, self.__simLogger)
self.__threadCamera.start()
self.__simLogger.warning("Reanimating Camera Thread")
except Exception as e:
self.__simLogger.critical(
"Some exception: " + str(e) + str(sys.exc_info()[0]) + " - " + traceback.format_exc()
)
self.__thymioController.stopThymioRequest()
break
self.__simLogger.info("End of while loop: " + str(evals) + " >= " + str(pr.total_evals))
self.stop()
self.__simLogger.info("_____END OF SIMULATION_____\n")
def isStopped(self):
return self.__stopped
def stop(self):
if self.__stopped:
self.__simLogger.info("Simulation already stopped.")
return
self.__stopped = True
# Send log files to server.
self.__simLogger.info("Sending Files...\n")
self.__sendFiles(
self.__simulationOutputFile,
self.__simulationLogFile,
self.__simulationTempFile,
self.__simulationWeightOutputFile,
)
# Stop Thymio from moving
self.__thymioController.stopThymioRequest()
# Stopping all the message senders: no more outgoing messages
for addr in self.__msgSenders:
self.__simLogger.info("Killing Sender " + addr)
self.__msgSenders[addr].stop()
self.__msgSenders[addr].join()
self.__simLogger.info("All MessageSenders: KILLED")
# Stopping connections listener: no more incoming connections
self.__connListener.stop()
self.__connListener.join()
self.__simLogger.info("ConnectionsListener: KILLED")
# Stopping camera thread
self.__threadCamera.stop()
self.__threadCamera.join()
self.__simLogger.info("CameraThread: KILLED")
# Stopping all the message receivers: no more incoming messages
i = 0
for addr in self.__msgReceivers:
self.__simLogger.info("Killing Receiver " + addr)
self.__msgReceivers[addr].stop()
sendOneMessage(self.__stopSockets[i], "STOP") # Send stop messages
self.__msgReceivers[addr].join()
self.__stopSockets[i].close()
i += 1
self.__simLogger.info("All MessageReceivers: KILLED")