# HDPy.puppy.PuppyHDP
tumbled_reward = 0.0,
# HDPy.ADHDP
reservoir = reservoir,
readout = readout,
# HDPy.ActorCritic
plant = plant,
policy = collector,
gamma = 0.5,
alpha = 1.0,
init_steps = 10,
norm = nrm
)
# robot
r = PuPy.robotBuilder(
Robot,
actor,
sampling_period_ms = 20,
ctrl_period_ms = 3000,
# event_handlers = actor.event_handler
)
## SIMULATION LOOP ##
# run the simulation
r.run()
# teardown
readout.save('/tmp/puppy_readout.pic')
# gait params
gait_params = {
"frequency": (1.0, 1.0, 1.0, 1.0),
"offset": (-0.23, -0.23, -0.37, -0.37),
"amplitude": (0.56, 0.56, 0.65, 0.65),
"phase": (0.0, 0.0, 0.5, 0.5),
}
# gaits
gait = PuPy.Gait(gait_params)
# Multidimensional collector
class MyCollector(PuPy.RobotCollector):
def __call__(self, epoch, time_start_ms, time_end_ms, step_size):
if len(epoch) > 0:
epoch["random"] = np.random.normal(size=(100, 10))
return super(MyCollector, self).__call__(epoch, time_start_ms, time_end_ms, step_size)
# actor
actor = PuPy.ConstantGaitControl(gait)
observer = MyCollector(actor, expfile="/tmp/puppy_sim.hdf5")
# robot
r = PuPy.robotBuilder(Robot, observer, sampling_period_ms=20, noise_ctrl=None, noise_obs=None)
# run
r.run()
from controller import Supervisor import PuPy # checks checks = [] checks.append(PuPy.RevertOutOfArena(arena_size=(-10, 10, -10, 10), distance=0, grace_time_ms=(3 * 3000))) # respawn the robot at a random location in a bounded area checks.append(PuPy.RevertTumbled(grace_time_ms=(3 * 3000))) # set up supervisor s = PuPy.supervisorBuilder(Supervisor, 20, checks) # run s.run()
while grp_name not in f and idx >= 0:
idx -= 1
grp_name = "traj_%03i" % (idx)
if idx < 0:
raise Exception("Could not find last trajectory")
trajectory = f[grp_name][:]
do_quit = True
f.close()
# Initialize the collector
collector = PuPy.RobotCollector(child=policy, expfile="/tmp/example_data.hdf5")
# Initialize the actor
actor = TrajectoryFollower(trajectory=trajectory, policy=collector, init_steps=10)
# Initialize the robot, bind it to webots
r = PuPy.robotBuilder(Robot, actor, sampling_period_ms=20, ctrl_period_ms=3000)
# Register robot in actor for signalling
actor.robot = r
if do_quit:
# Quit the simulation when all trajectories are handled
r.send_msg("quit_on_demand")
# Run the simulation
r.run()
from controller import Supervisor
import PuPy
# emitter checks
class EmitterCheck(PuPy.SupervisorCheck):
def __call__(self, supervisor):
supervisor.emitter.send('Emitting ' + str(supervisor.num_iter))
checks = []
checks.append(EmitterCheck())
# set up supervisor
PuPy.supervisorBuilder(Supervisor, 100, checks).run()
from controller import Supervisor import PuPy # checks checks = [PuPy.QuitOnDemand(), PuPy.RevertOnDemand()] # set up supervisor s = PuPy.supervisorBuilder(Supervisor, 20, [PuPy.ReceiverCheck(checks)]) # run s.run()
tumble_collector = PuPy.TumbleCollector(
child = data_collector,
sampling_period_ms = sampling_period_ms,
ctrl_period_ms = ctrl_period_ms
)
collector = PuPy.ResetCollector(
child = tumble_collector,
sampling_period_ms = sampling_period_ms,
ctrl_period_ms = ctrl_period_ms
)
# actor instantiation
actor = OfflinePuppy(
# policy = collector,
policy = data_collector,
init_steps = 10,
)
# robot instantiation
r = PuPy.robotBuilder(
Robot,
actor,
sampling_period_ms = sampling_period_ms,
ctrl_period_ms = ctrl_period_ms,
# event_handlers = [actor.event_handler, tumble_collector.event_handler, collector.event_handler]
)
# invoke the main loop, starts the simulation
r.run()
import os.path
import numpy as np
# gait params
gait_params = {
'frequency' : (1.0, 1.0, 1.0, 1.0),
'offset' : ( -0.23, -0.23, -0.37, -0.37),
'amplitude' : ( 0.56, 0.56, 0.65, 0.65),
'phase' : (0.0, 0.0, 0.5, 0.5)
}
# gaits
gait = PuPy.Gait(gait_params)
# receiver callback
def recv1(robot, epoch, current_time, msg):
print "recv1", msg
def recv2(robot, epoch, current_time, msg):
print "recv2", msg
# actor
actor = PuPy.ConstantGaitControl(gait)
# robot
r = PuPy.robotBuilder(Robot, actor, event_period_ms=50, event_handlers=[recv1, recv2])
r.run()
