def __init__(self,
target_position,
target_rotation,
skip_frame=20,
model_path='rubik_cube_shadow_wo_target.xml',
initial_robot_pos_dict=DEFAULT_INITIAL_QPOS,
reward_type='sparse',
randomize_initial_position=True,
randomize_initial_rotation=True,
distance_threshold=0.01,
rotation_threshold=0.1,
target_position_range=np.array([(-0.04, 0.04), (-0.06, 0.02),
(0.0, 0.06)])):
"""Initializes a new ManipulateEnv environment.
Args:
model_path (string): path to the environment XML file
initial_robot_pos_dict (dict): a dictionary of joint names and values that define the initial configuration
target_position (string): the type of target porisiton:
- ignore: target position is fully ignored, i.e. the object can be positioned arbitrarily
- fixed: target position is set to the initial position of the target
- random: target position is fully randomized according to target_postion_range
target_rotation (string): the type of target rotation:
- ignore: target rotation is fully ignored, i.e. the object can be rotated arbitrarily
- fixed: target rotation is set to the initial rotation of the object
- xyz: fully randomized target rotation around the X, Y and Z axis
- z: fully randomized target rotation around the Z axis
- parallel: fully randomized target rotation around Z and axis-aligned rotation around X, Y
target_position_range (np.array of shape (3,2)): range of the target_position randomization
reward_type ('sparse' or 'dense'): the reward type, i.e. sparse or dense
randomize_initial_position (boolean): whether or not to randomize the initial position of the object
randomize_initial_rotation (boolean): whether or not to randomize the initial rotation of the object
distance_threshold (float, in meters): the threshold after which the position of a goal is considered achieved
rotation_threshold (float, in radians): the threshold after which the rotation of a goal is considered achieved
"""
self.target_position = target_position
self.target_rotation = target_rotation
self.target_position_range = target_position_range
self.reward_type = reward_type
self.randomize_initial_position = randomize_initial_position
self.randomize_initial_rotation = randomize_initial_rotation
self.distance_threshold = distance_threshold
self.rotation_threshold = rotation_threshold
self.parallel_quats = [
rotations.euler2quat(r)
for r in rotations.get_parallel_rotations()
]
assert self.reward_type in ['dense', 'sparse']
assert self.target_position in ['ignore', 'fixed', 'random']
assert self.target_rotation in [
'ignore', 'fixed', 'xyz', 'z', 'parallel', 'rotFace'
]
ShadowGoalBaseEnv.__init__(self, model_path, initial_robot_pos_dict,
skip_frame)
def __init__(
self, model_path, target_position, target_rotation,
target_position_range, reward_type, initial_qpos={},
randomize_initial_position=True, randomize_initial_rotation=True,
distance_threshold=0.01, rotation_threshold=0.1, n_substeps=20, relative_control=False,
ignore_z_target_rotation=False,
):
"""Initializes a new Hand manipulation environment.
Args:
model_path (string): path to the environments XML file
target_position (string): the type of target position:
- ignore: target position is fully ignored, i.e. the object can be positioned arbitrarily
- fixed: target position is set to the initial position of the object
- random: target position is fully randomized according to target_position_range
target_rotation (string): the type of target rotation:
- ignore: target rotation is fully ignored, i.e. the object can be rotated arbitrarily
- fixed: target rotation is set to the initial rotation of the object
- xyz: fully randomized target rotation around the X, Y and Z axis
- z: fully randomized target rotation around the Z axis
- parallel: fully randomized target rotation around Z and axis-aligned rotation around X, Y
ignore_z_target_rotation (boolean): whether or not the Z axis of the target rotation is ignored
target_position_range (np.array of shape (3, 2)): range of the target_position randomization
reward_type ('sparse' or 'dense'): the reward type, i.e. sparse or dense
initial_qpos (dict): a dictionary of joint names and values that define the initial configuration
randomize_initial_position (boolean): whether or not to randomize the initial position of the object
randomize_initial_rotation (boolean): whether or not to randomize the initial rotation of the object
distance_threshold (float, in meters): the threshold after which the position of a goal is considered achieved
rotation_threshold (float, in radians): the threshold after which the rotation of a goal is considered achieved
n_substeps (int): number of substeps the simulation runs on every call to step
relative_control (boolean): whether or not the hand is actuated in absolute joint positions or relative to the current state
"""
self.target_position = target_position
self.target_rotation = target_rotation
self.target_position_range = target_position_range
self.parallel_quats = [rotations.euler2quat(r) for r in rotations.get_parallel_rotations()]
self.randomize_initial_rotation = randomize_initial_rotation
self.randomize_initial_position = randomize_initial_position
self.distance_threshold = distance_threshold
self.rotation_threshold = rotation_threshold
self.reward_type = reward_type
self.ignore_z_target_rotation = ignore_z_target_rotation
assert self.target_position in ['ignore', 'fixed', 'random']
assert self.target_rotation in ['ignore', 'fixed', 'xyz', 'z', 'parallel']
hand_env.HandEnv.__init__(
self, model_path, n_substeps=n_substeps, initial_qpos=initial_qpos,
relative_control=relative_control)
utils.EzPickle.__init__(self)
def __init__(
self, model_path, target_position, target_rotation,
target_position_range, reward_type, initial_qpos=None,
randomize_initial_position=True, randomize_initial_rotation=True,
distance_threshold=0.01, rotation_threshold=0.1, n_substeps=20, relative_control=False,
ignore_z_target_rotation=False,
):
"""Initializes a new Hand manipulation environment.
Args:
model_path (string): path to the environments XML file
target_position (string): the type of target position:
- ignore: target position is fully ignored, i.e. the object can be positioned arbitrarily
- fixed: target position is set to the initial position of the object
- random: target position is fully randomized according to target_position_range
target_rotation (string): the type of target rotation:
- ignore: target rotation is fully ignored, i.e. the object can be rotated arbitrarily
- fixed: target rotation is set to the initial rotation of the object
- xyz: fully randomized target rotation around the X, Y and Z axis
- z: fully randomized target rotation around the Z axis
- parallel: fully randomized target rotation around Z and axis-aligned rotation around X, Y
ignore_z_target_rotation (boolean): whether or not the Z axis of the target rotation is ignored
target_position_range (np.array of shape (3, 2)): range of the target_position randomization
reward_type ('sparse' or 'dense'): the reward type, i.e. sparse or dense
initial_qpos (dict): a dictionary of joint names and values that define the initial configuration
randomize_initial_position (boolean): whether or not to randomize the initial position of the object
randomize_initial_rotation (boolean): whether or not to randomize the initial rotation of the object
distance_threshold (float, in meters): the threshold after which the position of a goal is considered achieved
rotation_threshold (float, in radians): the threshold after which the rotation of a goal is considered achieved
n_substeps (int): number of substeps the simulation runs on every call to step
relative_control (boolean): whether or not the hand is actuated in absolute joint positions or relative to the current state
"""
self.target_position = target_position
self.target_rotation = target_rotation
self.target_position_range = target_position_range
self.parallel_quats = [rotations.euler2quat(r) for r in rotations.get_parallel_rotations()]
self.randomize_initial_rotation = randomize_initial_rotation
self.randomize_initial_position = randomize_initial_position
self.distance_threshold = distance_threshold
self.rotation_threshold = rotation_threshold
self.reward_type = reward_type
self.ignore_z_target_rotation = ignore_z_target_rotation
assert self.target_position in ['ignore', 'fixed', 'random']
assert self.target_rotation in ['ignore', 'fixed', 'xyz', 'z', 'parallel']
initial_qpos = initial_qpos or {}
hand_env.HandEnv.__init__(
self, model_path, n_substeps=n_substeps, initial_qpos=initial_qpos,
relative_control=relative_control)