def reset(self, reset_args=None, **kwargs):
if reset_args is None:
logger.log("Debug, warning, reset_args for env is None")
goal = reset_args
if goal is not None:
assert len(goal) == 2, "wrong size goal"
goal_num, test = goal
if (goal_num != self.goal_num) or (test != self.test):
if self.mujoco.viewer is not None:
self.mujoco.stop_viewer()
self.mujoco.terminate()
self.goal_num, self.test = goal
demo_path = (self.train_dir + str(self.goal_num) +
".pkl") if not self.test else (
self.test_dir + str(self.goal_num) + ".pkl")
demo_data = joblib.load(demo_path)
xml_file = demo_data["xml"]
xml_file = xml_file.replace(
"/root/code/rllab/vendor/mujoco_models/", self.xml_dir)
# print("debug,xml_file", xml_file)
if int(xml_file[-5]) % 2 == 0 and not self.debug:
print("inverted_order", xml_file, self.goal_num, self.test)
self.shuffle_order = [1, 0]
else:
print("normal_order", xml_file, self.goal_num, self.test)
self.shuffle_order = [0, 1]
self.mujoco = mujoco_env.MujocoEnv(file_path=xml_file)
# else:
# print("continuing with xml", self.mujoco.FILE, self.goal_num, self.test)
elif self.goal_num is None: #if we already have a goal_num, we don't sample a new one, just reset the model
self.goal_num, self.test = self.sample_goals(num_goals=1,
test=False)[0]
demo_path = (self.train_dir + str(self.goal_num) +
".pkl") if not self.test else (self.test_dir +
str(self.goal_num) +
".pkl")
demo_data = joblib.load(demo_path)
xml_file = demo_data["xml"]
xml_file = xml_file.replace(
"/root/code/rllab/vendor/mujoco_models/", self.xml_dir)
if int(xml_file[-5]) % 2 == 0 and not self.debug:
print("inverted_order, first time initializing env", xml_file)
self.shuffle_order = [
1, 0
] # TODO: flip back, this is set to [0,1] just for debugging purposes
else:
print("normal_order, first time initializing env", xml_file)
self.shuffle_order = [0, 1]
self.mujoco = mujoco_env.MujocoEnv(file_path=xml_file)
# self.viewer_setup()
self.reset_model()
return self.get_current_obs()
def reset(self, reset_args=None, **kwargs):
# print("debug,asked to reset with reset_args", reset_args)
qpos = np.copy(self.mujoco.init_qpos)
qvel = np.copy(
self.mujoco.init_qvel) + 0.0 * self.mujoco.np_random.uniform(
low=-0.005, high=0.005, size=self.mujoco.model.nv)
goal_num = reset_args
if goal_num is not None:
if self.goal_num != goal_num:
self.goal_num = goal_num
self.shuffle_order = [[0, 1, 2], [1, 2, 0], [2, 0,
1]][self.goal_num]
# if self.mujoco.viewer is not None:
# self.mujoco.stop_viewer()
# self.mujoco.release()
# self.mujoco.terminate()
# self.mujoco.terminate()
xml_file = '/home/kevin/maml_rl/vendor/mujoco_models/r7dof_versions/reacher_7dof_2distr_%s%s%s.xml' % tuple(
self.shuffle_order)
self.mujoco = mujoco_env.MujocoEnv(file_path=xml_file)
self.viewer_setup()
elif self.goal_num is None: # do not change color of goal or XML file between resets.
self.goal_num = np.random.choice([0, 1, 2])
self.shuffle_order = [[0, 1, 2], [1, 2, 0], [2, 0,
1]][self.goal_num]
if self.mujoco.viewer is not None:
self.mujoco.stop_viewer()
self.mujoco.release()
self.mujoco.terminate()
xml_file = '/home/kevin/maml_rl/vendor/mujoco_models/r7dof_versions/reacher_7dof_2distr_%s%s%s.xml' % tuple(
self.shuffle_order)
self.mujoco = mujoco_env.MujocoEnv(file_path=xml_file)
self.viewer_setup()
self.goal = np.random.uniform(low=[-0.4, -0.4, -0.3],
high=[0.4, 0.0, -0.3]).reshape(3, 1)
self.distract1 = np.random.uniform(low=[-0.4, -0.4, -0.3],
high=[0.4, 0.0,
-0.3]).reshape(3, 1)
self.distract2 = np.random.uniform(low=[-0.4, -0.4, -0.3],
high=[0.4, 0.0,
-0.3]).reshape(3, 1)
qpos[-14:-11] = self.distract1
qpos[-21:-18] = self.distract2
qpos[-7:-4] = self.goal
qvel[-7:] = 0
setattr(self.mujoco.model.data, 'qpos', qpos)
setattr(self.mujoco.model.data, 'qvel', qvel)
self.mujoco.model.data.qvel = qvel
self.mujoco.model._compute_subtree()
self.mujoco.model.forward()
self.current_com = self.mujoco.model.data.com_subtree[0]
self.dcom = np.zeros_like(self.current_com)
return self.get_current_image_obs()[1]
def __init__(self,
xml_file=None,
distance_metric_order=None,
distractors=True,
*args,
**kwargs):
self.goal = None
if 'noise' in kwargs:
noise = kwargs['noise']
else:
noise = 0.0
self.include_distractors = distractors
if self.include_distractors:
self.shuffle_order = [[0, 1, 2], [1, 2, 0], [2, 0, 1]][2]
# self.shuffle_order = rd.sample([[0,1,2],[1,2,0],[2,0,1]],1)[0]
if xml_file is None:
if not self.include_distractors:
xml_file = '/home/kevin/maml_rl/vendor/mujoco_models/r7dof_versions/reacher_7dof.xml'
else:
xml_file = '/home/kevin/maml_rl/vendor/mujoco_models/r7dof_versions/reacher_7dof_2distr_%s%s%s.xml' % tuple(
self.shuffle_order)
print("xml file", xml_file)
self.mujoco = mujoco_env.MujocoEnv(file_path=xml_file,
action_noise=noise)
self.action_space = self.mujoco.action_space
self.get_viewer = self.mujoco.get_viewer
self.log_diagnostics = self.mujoco.log_diagnostics
Serializable.__init__(self, *args, **kwargs)
def reset(self, reset_args=None, **kwargs):
goal = reset_args
if goal is not None:
assert len(goal) == 2, "wrong size goal"
goal_num, test = goal
if (goal_num != self.goal_num) or (test != self.test):
if self.viewer is not None:
self.stop_viewer()
self.terminate()
self.goal_num, self.test = goal
demo_path = (self.train_dir + str(self.goal_num) +
".pkl") if not self.test else (
self.test_dir + str(self.goal_num) + ".pkl")
demo_data = joblib.load(demo_path)
xml_file = demo_data["xml"]
xml_file = xml_file.replace(
"/root/code/rllab/vendor/mujoco_models/", self.xml_dir)
print("debug,xml_file", xml_file)
if int(xml_file[-5]) % 2 == 0:
print("retaining order")
self.shuffle_order = [0, 1]
else:
print("flipping order")
self.shuffle_order = [1, 0]
self = mujoco_env.MujocoEnv(file_path=xml_file)
elif self.goal_num is None: #if we already have a goal_num, we don't sample a new one, just reset the model
self.goal_num, self.test = self.sample_goals(num_goals=1,
test=False)[0]
demo_path = (self.train_dir + str(self.goal_num) +
".pkl") if not self.test else (self.test_dir +
str(self.goal_num) +
".pkl")
demo_data = joblib.load(demo_path)
xml_file = demo_data["xml"]
xml_file = xml_file.replace(
"/root/code/rllab/vendor/mujoco_models/", self.xml_dir)
if int(xml_file[-5]) % 2 == 0:
print("retaining order")
self.shuffle_order = [0, 1]
else:
print("flipping order")
self.shuffle_order = [1, 0]
self = mujoco_env.MujocoEnv(file_path=xml_file)
self.viewer_setup()
self.reset_model()
return self.get_current_obs()
def __init__(self,
xml_file=None,
goal_num=None,
distance_metric_order=None,
distractors=True,
*args,
**kwargs):
self.goal_num = np.random.choice([0, 1, 2])
self.shuffle_order = [[0, 1, 2], [1, 2, 0], [2, 0, 1]][self.goal_num]
self.include_distractors = distractors
assert distractors == True, "not supported"
if xml_file is None:
xml_file = '/home/kevin/maml_rl/vendor/mujoco_models/r7dof_versions/reacher_7dof_2distr_%s%s%s.xml' % tuple(
self.shuffle_order)
print("xml file", xml_file)
self.mujoco = mujoco_env.MujocoEnv(file_path=xml_file)
self.viewer_setup()
self.action_space = self.mujoco.action_space
self.get_viewer = self.mujoco.get_viewer
self.log_diagnostics = self.mujoco.log_diagnostics
Serializable.__init__(self, *args, **kwargs)