loco_mujoco.utils package

loco_mujoco.utils.domain_randomization module

loco_mujoco.utils.goals module

class GoalDirectionVelocity

Bases: object

get_direction()

get_goal()

get_velocity()

set_goal(direction, velocity)

Rewards

class CustomReward(reward_callback=None)

Bases: RewardInterface

class MultiTargetVelocityReward(target_velocity, x_vel_idx, env_id_len, scalings)

Bases: RewardInterface

class NoReward

Bases: RewardInterface

A reward function that returns always 0.

class PosReward(pos_idx)

Bases: RewardInterface

class RewardInterface

Bases: object

Interface to specify a reward function.

reset_state()

Reset the state of the object.

class TargetVelocityReward(target_velocity, x_vel_idx)

Bases: RewardInterface

class VelocityVectorReward(x_vel_idx, y_vel_idx, angle_idx, goal_vel_idx)

Bases: RewardInterface

Trajectory

class Trajectory(keys, low, high, joint_pos_idx, interpolate_map, interpolate_remap, traj_path=None, traj_files=None, interpolate_map_params=None, interpolate_remap_params=None, traj_dt=0.002, control_dt=0.01, ignore_keys=None, clip_trajectory_to_joint_ranges=False, traj_info=None, warn=True)

Bases: object

General class to handle trajectory data. It builds a general trajectory from a numpy bin file(.npy), and automatically interpolates the trajectory to the desired control frequency. This class is used to generate datasets and to sample from the dataset to initialize the simulation.

All trajectories are required to be of equal length.

check_if_trajectory_is_in_range(low, high, keys, j_idx, warn, clip_trajectory_to_joint_ranges)

create_dataset(ignore_keys=None, state_callback=None, state_callback_params=None)

Creates a dataset used by imitation learning algorithms.

Args:

ignore_keys (list): List of keys to ignore in the dataset. state_callback (func): Function that should be called on each state. state_callback_params (dict): Dictionary of parameters needed to make

the state transformation.

Returns:

Dictionary containing states, next_states, absorbing and last flags. For the states the shape is (N_traj x N_samples_per_traj-1, dim_state), while the flags have the shape (N_traj x N_samples_per_traj-1). If traj_info was specified, it will also include that.

flattened_trajectories()

Returns the trajectories flattened in the N_traj dimension. Also expands dim if obs has dimension 1.

get_current_sample()

Returns the current sample in the trajectory.

get_from_sample(sample, key)

Returns the part of the sample whose key is specified. In contrast to the function _get_from_obs from the base environment, this function also allows to access information that is in the trajectory, but not in the simulation such as goal definitions.

Note: This function is not suited for getting an observation from environment samples!

Args:

sample (list or np.array): Current sample to extract an observation from. key (string): Name of the observation to extract from sample

Returns:

np.array consisting of the observation specified by the key.

get_idx(key)

Returns the index of the key.

Note: This function is not suited for getting the index for an observation of the environment!

Args:

key (string): Name of the observation to extract from sample

Returns:

int containing the desired index.

get_next_sample()

Returns the next sample in the trajectory.

reset_trajectory(substep_no=None, traj_no=None)

Resets the trajectory to a certain trajectory and a substep within that trajectory. If one of them is None, they are set randomly.

Args:

substep_no (int, None): Starting point of the trajectory.

If None, the trajectory starts from a random point.

traj_no (int, None): Number of the trajectory to start from.

If None, it starts from a random trajectory

Returns:

The chosen (or randomly sampled) sample from a trajectory.

property number_obs_trajectory

Returns the number of observations in the trajectory.

property number_of_trajectories

Returns the number of trajectories.

property trajectory_length

Returns the length of a trajectory. Note that all trajectories have to be equal in length.