def viewer_setup(self):
HopperEnv.viewer_setup(self)
self.viewer.cam.trackbodyid = 2
self.viewer.cam.distance = 8.0
self.viewer.cam.lookat[2] = 1.15
self.viewer.cam.elevation = 0
mj.functions.mjr_uploadHField(self.model,
self.sim.render_contexts[0].con, 0)
def __init__(self, gravity=-9.81, *args, **kwargs):
HopperEnv.__init__(self)
utils.EzPickle.__init__(self)
# make sure we're using a proper OpenAI gym Mujoco Env
assert isinstance(self, mujoco_env.MujocoEnv)
self.model.opt.gravity = (mujoco_py.mjtypes.c_double *
3)(*[0., 0., gravity])
self.model._compute_subtree()
self.model.forward()
def __init__(
self,
gravity=-9.81,
*args,
**kwargs):
HopperEnv.__init__(self)
utils.EzPickle.__init__(self)
# make sure we're using a proper OpenAI gym Mujoco Env
assert isinstance(self, mujoco_env.MujocoEnv)
self.model.opt.gravity = (mujoco_py.mjtypes.c_double * 3)(*[0., 0., gravity])
self.model._compute_subtree()
self.model.forward()
def _get_obs(self):
obs = np.concatenate([
HopperEnv._get_obs(self),
np.zeros(self.n_bins)
# goal_readings
])
return obs
def _get_obs(self):
obs = np.concatenate([
HopperEnv._get_obs(self),
np.zeros(self.n_bins)
# goal_readings
])
return obs
def change_robot_body(self, params):
"""Changes the robot body parameters accordingly to the passed in dictionary.
Args:
params: dictionary containing the parameters for changing the robot body.
"""
# Check to make sure params are all within correct range
for key in params.keys():
if 'rgb' in key:
if len(params[key]) != 3:
print('Color parameter ' + key +
' must be list of length 3')
return
if any([value > 1 or value < 0 for value in params[key]]):
print('RGB color values must be between 0 and 1.')
return
else:
if params[key] < 0.5 or params[key] > 2.0:
print('Scaling value ' + key +
' must be between 0.5 and 2.0')
return
if self.env:
self.env.close()
if 'cheetah' in self.robot_name:
# generate random filename for xml
xml = 'half_cheetah_' + ''.join(
random.choice(string.ascii_lowercase + string.digits)
for i in range(8)) + '.xml'
cheetah(xml_name=xml, **params)
if self.env:
self.env.close()
self.env = HalfCheetahEnv(xml)
elif 'ant' in self.robot_name:
# generate random filename for xml
xml = 'ant_' + ''.join(
random.choice(string.ascii_lowercase + string.digits)
for i in range(8)) + '.xml'
ant(xml_name=xml, **params)
if self.env:
self.env.close()
self.env = AntEnv(xml)
elif 'hopper' in self.robot_name:
# generate random filename for xml
xml = 'hopper_' + ''.join(
random.choice(string.ascii_lowercase + string.digits)
for i in range(8)) + '.xml'
hopper(xml_name=xml, **params)
self.env = HopperEnv(xml)
elif 'walker' in self.robot_name:
# generate random filename for xml
xml = 'walker_' + ''.join(
random.choice(string.ascii_lowercase + string.digits)
for i in range(8)) + '.xml'
walker(xml_name=xml, **params)
self.env = Walker2dEnv(xml)
else:
print(
"Illegal Robot Name. Must contain 'cheetah', 'ant', 'hopper', or 'walker'"
)
class CustomRobot(Robot):
"""CustomRobot class for designing and creating simulated robots.
"""
def __init__(self, robot_name, params=None, linear=None):
"""Initialize the robot.
Args:
robot_name: name of the robot (e.g. hopper, ant, walker, cheetah)
params: dictionary of robot body parameters. if None, use defaults.
linear: if True, use a linear policy. if False, use a neural network.
"""
self.robot_name = robot_name.lower()
self.env = None
self.linear = linear
if params is None:
params = self.get_default_body_params()
# change_robot_body initializes the env
self.change_robot_body(params)
self.max_timesteps = 500
self.action_dim = self.env.action_space.shape[0]
super(CustomRobot, self).__init__(self.env)
def get_default_body_params(self):
"""Returns dictionary of default body parameters. Can be modified and passed to
change_robot_body to modify the body of the robot.
"""
params = {}
if 'cheetah' in self.robot_name:
# Color of body and feet, expressed as RGB intensities
params['body_rgb'] = (0.8, 0.6, 0.4)
params['feet_rgb'] = (0.9, 0.6, 0.6)
# Scaling factor for different body parts.
param_names = [
'torso', 'head', 'back_foot', 'front_foot', 'back_thigh',
'front_thigh', 'back_shin', 'front_shin', 'limb_width'
]
for key in param_names:
params[key] = 1.0
elif 'ant' in self.robot_name:
# Color of body and feet, expressed as RGB intensities
params['torso_rgb'] = (0.8, 0.6, 0.4)
params['legs_rgb'] = (0.8, 0.6, 0.4)
# Scaling factor for different body parts.
param_names = [
'torso', 'front_hips', 'back_hips', 'front_legs', 'back_legs',
'front_feet', 'back_feet', 'limb_width'
]
for key in param_names:
params[key] = 1.0
elif 'hopper' in self.robot_name:
# Color of body and feet, expressed as RGB intensities
params['torso_rgb'] = (0.8, 0.6, 0.4)
params['leg_rgb'] = (0.7, 0.3, 0.6)
# Scaling factor for different body parts.
param_names = ['torso', 'thigh', 'leg', 'foot', 'limb_width']
for key in param_names:
params[key] = 1.0
elif 'walker' in self.robot_name:
# Color of body and feet, expressed as RGB intensities
params['torso_rgb'] = (0.8, 0.6, 0.4)
params['right_leg_rgb'] = (0.8, 0.6, 0.4)
params['left_leg_rgb'] = (0.7, 0.3, 0.6)
# Scaling factor for different body parts.
param_names = ['torso', 'thighs', 'legs', 'feet', 'limb_width']
for key in param_names:
params[key] = 1.0
return params
def change_robot_body(self, params):
"""Changes the robot body parameters accordingly to the passed in dictionary.
Args:
params: dictionary containing the parameters for changing the robot body.
"""
# Check to make sure params are all within correct range
for key in params.keys():
if 'rgb' in key:
if len(params[key]) != 3:
print('Color parameter ' + key +
' must be list of length 3')
return
if any([value > 1 or value < 0 for value in params[key]]):
print('RGB color values must be between 0 and 1.')
return
else:
if params[key] < 0.5 or params[key] > 2.0:
print('Scaling value ' + key +
' must be between 0.5 and 2.0')
return
if self.env:
self.env.close()
if 'cheetah' in self.robot_name:
# generate random filename for xml
xml = 'half_cheetah_' + ''.join(
random.choice(string.ascii_lowercase + string.digits)
for i in range(8)) + '.xml'
cheetah(xml_name=xml, **params)
if self.env:
self.env.close()
self.env = HalfCheetahEnv(xml)
elif 'ant' in self.robot_name:
# generate random filename for xml
xml = 'ant_' + ''.join(
random.choice(string.ascii_lowercase + string.digits)
for i in range(8)) + '.xml'
ant(xml_name=xml, **params)
if self.env:
self.env.close()
self.env = AntEnv(xml)
elif 'hopper' in self.robot_name:
# generate random filename for xml
xml = 'hopper_' + ''.join(
random.choice(string.ascii_lowercase + string.digits)
for i in range(8)) + '.xml'
hopper(xml_name=xml, **params)
self.env = HopperEnv(xml)
elif 'walker' in self.robot_name:
# generate random filename for xml
xml = 'walker_' + ''.join(
random.choice(string.ascii_lowercase + string.digits)
for i in range(8)) + '.xml'
walker(xml_name=xml, **params)
self.env = Walker2dEnv(xml)
else:
print(
"Illegal Robot Name. Must contain 'cheetah', 'ant', 'hopper', or 'walker'"
)
def get_action_size(self):
""" Returns the dimensionality of the robot's action space.
"""
return self.action_dim
def run(self, actions=None, action_durations=None):
""" Runs the robot in the environment and returns a video clip.
Args:
actions:
if actions is None, use random actions.
if actions is a list of scalar numbers, execute the specified action continuously.
The length of the list must be the number of joints of the robot. Get the
number of joints by calling get_action_size()
if actions is a list of lists and action_durations is a list,
periodically execute each action in the list actions for the duration
specified in action_durations. The length of actions and action_durations
should be the same.
"""
if actions is not None and action_durations is None:
if len(actions) != self.action_dim:
print('Must specify an action of size ' + str(self.action_dim))
print('Size of passed in action is ' + str(len(actions)))
return
# With a constant action, only simulate for a short amount of time.
max_timesteps = 100
else:
max_timesteps = self.max_timesteps
if actions is not None and action_durations is not None:
if len(actions) != len(action_durations):
print(
'Number of actions must be the same as the number of action_durations.'
)
print('len(actions) is ' + str(len(actions)))
print('len(action_durations) is ' + str(len(action_durations)))
return
cur_action_id = 0
cur_action_timestep = 0
print('Running.')
obs = self.env.reset()
video = []
for t in range(max_timesteps):
if t % 2 == 0:
video.append(self.get_image())
if actions == None:
# Sample random actions.
action = 0.1 * np.random.normal(size=(self.action_dim, 1))
elif actions is not None and action_durations == None:
action = actions
else:
action = actions[cur_action_id]
cur_action_timestep += 1
if cur_action_timestep > action_durations[cur_action_id]:
cur_action_id += 1
cur_action_timestep = 0
if cur_action_id >= len(actions):
cur_action_id = 0
obs, _, _, _ = self.env.step(action)
center_of_mass = self.env.get_body_com("torso")[0]
if center_of_mass > 0:
print('Done running. Your robot went ' +
'{0:.2f}'.format(center_of_mass) + ' meters forward.')
else:
print('Done running. Your robot went ' +
'{0:.2f}'.format(abs(center_of_mass)) + ' meters backward.')
video_clip = mpy.ImageSequenceClip(video, fps=20)
return video_clip