toddlerbot.locomotion package

Submodules

toddlerbot.locomotion.mjx_config module

class toddlerbot.locomotion.mjx_config.MJXConfig

Bases: object

Configuration class for the MJX environment.

class ActionConfig(action_parts: List[str] = <factory>, action_scale: float = 0.25, filter_type: str = 'none', filter_order: int = 4, filter_cutoff: float = 10.0, contact_force_threshold: float = 1.0, n_steps_delay: int = 1, n_frames: int = 5, cycle_time: float = 0.72, waist_roll_max: float = 0.0)

Bases: object

action_parts: List[str]

action_scale: float = 0.25

contact_force_threshold: float = 1.0

cycle_time: float = 0.72

filter_cutoff: float = 10.0

filter_order: int = 4

filter_type: str = 'none'

n_frames: int = 5

n_steps_delay: int = 1

waist_roll_max: float = 0.0

class CommandsConfig(resample_time: float = 3.0, reset_time: float = 100.0, mean_reversion: float = 0.5, zero_chance: float = 0.2, turn_chance: float = 0.3, command_obs_indices: List[int] = <factory>, command_range: List[List[float]] = <factory>, deadzone: List[float] = <factory>)

Bases: object

command_obs_indices: List[int]

command_range: List[List[float]]

deadzone: List[float]

mean_reversion: float = 0.5

resample_time: float = 3.0

reset_time: float = 100.0

turn_chance: float = 0.3

zero_chance: float = 0.2

class DomainRandConfig(add_domain_rand: bool = True, friction_range: List[float] = <factory>, damping_range: List[float] = <factory>, armature_range: List[float] = <factory>, frictionloss_range: List[float] = <factory>, body_mass_range: List[float] = <factory>, ee_mass_range: List[float] = <factory>, other_mass_range: List[float] = <factory>, kp_range: List[float] = <factory>, kd_range: List[float] = <factory>, tau_max_range: List[float] = <factory>, q_dot_tau_max_range: List[float] = <factory>, q_dot_max_range: List[float] = <factory>, push_interval_s: int = 2, push_lin_vel: float = 0.5, push_ang_vel: float = 1.0)

Bases: object

add_domain_rand: bool = True

armature_range: List[float]

body_mass_range: List[float]

damping_range: List[float]

ee_mass_range: List[float]

friction_range: List[float]

frictionloss_range: List[float]

kd_range: List[float]

kp_range: List[float]

other_mass_range: List[float]

push_ang_vel: float = 1.0

push_interval_s: int = 2

push_lin_vel: float = 0.5

q_dot_max_range: List[float]

q_dot_tau_max_range: List[float]

tau_max_range: List[float]

class NoiseConfig(add_noise: bool = True, action_noise: float = 0.02, obs_noise_scale: float = 0.05, dof_pos: float = 1.0, dof_vel: float = 2.0, ang_vel: float = 5.0, euler: float = 2.0, backlash_scale: float = 0.02, backlash_activation: float = 0.1)

Bases: object

action_noise: float = 0.02

add_noise: bool = True

ang_vel: float = 5.0

backlash_activation: float = 0.1

backlash_scale: float = 0.02

dof_pos: float = 1.0

dof_vel: float = 2.0

euler: float = 2.0

obs_noise_scale: float = 0.05

class ObsConfig(frame_stack: int = 15, c_frame_stack: int = 15, num_single_obs: int = 83, num_single_privileged_obs: int = 126)

Bases: object

c_frame_stack: int = 15

frame_stack: int = 15

num_single_obs: int = 83

num_single_privileged_obs: int = 126

class ObsScales(lin_vel: float = 2.0, ang_vel: float = 1.0, dof_pos: float = 1.0, dof_vel: float = 0.05, euler: float = 1.0)

Bases: object

ang_vel: float = 1.0

dof_pos: float = 1.0

dof_vel: float = 0.05

euler: float = 1.0

lin_vel: float = 2.0

class RewardScales(torso_pos: float = 0.0, torso_quat: float = 0.0, torso_roll: float = 0.0, torso_pitch: float = 0.0, lin_vel_xy: float = 1.0, lin_vel_z: float = 1.0, ang_vel_xy: float = 1.0, ang_vel_z: float = 1.0, neck_motor_pos: float = 0.0, arm_motor_pos: float = 0.0, waist_motor_pos: float = 0.0, leg_motor_pos: float = 5.0, motor_torque: float = 0.01, energy: float = 0.01, neck_action_rate: float = 0.0, neck_action_acc: float = 0.0, arm_action_rate: float = 0.0, arm_action_acc: float = 0.0, waist_action_rate: float = 0.0, waist_action_acc: float = 0.0, leg_action_rate: float = 0.05, leg_action_acc: float = 0.05, feet_contact: float = 1.0, collision: float = 0.0, survival: float = 10.0, feet_air_time: float = 0.0, feet_distance: float = 0.0, feet_slip: float = 0.0, feet_clearance: float = 0.0, stand_still: float = 0.0, align_ground: float = 0.0)

Bases: object

align_ground: float = 0.0

ang_vel_xy: float = 1.0

ang_vel_z: float = 1.0

arm_action_acc: float = 0.0

arm_action_rate: float = 0.0

arm_motor_pos: float = 0.0

collision: float = 0.0

energy: float = 0.01

feet_air_time: float = 0.0

feet_clearance: float = 0.0

feet_contact: float = 1.0

feet_distance: float = 0.0

feet_slip: float = 0.0

leg_action_acc: float = 0.05

leg_action_rate: float = 0.05

leg_motor_pos: float = 5.0

lin_vel_xy: float = 1.0

lin_vel_z: float = 1.0

motor_torque: float = 0.01

neck_action_acc: float = 0.0

neck_action_rate: float = 0.0

neck_motor_pos: float = 0.0

reset()

stand_still: float = 0.0

survival: float = 10.0

torso_pitch: float = 0.0

torso_pos: float = 0.0

torso_quat: float = 0.0

torso_roll: float = 0.0

waist_action_acc: float = 0.0

waist_action_rate: float = 0.0

waist_motor_pos: float = 0.0

class RewardsConfig(healthy_z_range: List[float] = <factory>, tracking_sigma: float = 100.0, min_feet_y_dist: float = 0.05, max_feet_y_dist: float = 0.13, torso_roll_range: List[float] = <factory>, torso_pitch_range: List[float] = <factory>)

Bases: object

healthy_z_range: List[float]

max_feet_y_dist: float = 0.13

min_feet_y_dist: float = 0.05

torso_pitch_range: List[float]

torso_roll_range: List[float]

tracking_sigma: float = 100.0

class SimConfig(timestep: float = 0.004, solver: int = 2, iterations: int = 1, ls_iterations: int = 4)

Bases: object

iterations: int = 1

ls_iterations: int = 4

solver: int = 2

timestep: float = 0.004

toddlerbot.locomotion.mjx_config.get_env_config(env: str)

Retrieves and parses the configuration for a specified environment.

Parameters:

env (str) – The name of the environment for which to retrieve the configuration.

Returns:

An instance of MJXConfig initialized with the parsed configuration.

Return type:

MJXConfig

Raises:

FileNotFoundError – If the configuration file for the specified environment does not exist.

toddlerbot.locomotion.mjx_env module

toddlerbot.locomotion.ppo_config module

class toddlerbot.locomotion.ppo_config.PPOConfig(policy_hidden_layer_sizes: Tuple[int, ...] = (512, 256, 128), value_hidden_layer_sizes: Tuple[int, ...] = (512, 256, 128), num_timesteps: int = 100000000, num_evals: int = 1000, episode_length: int = 1000, unroll_length: int = 20, num_updates_per_batch: int = 4, discounting: float = 0.97, learning_rate: float = 0.0001, decay_steps: int = 50000000, alpha: float = 0.1, entropy_cost: float = 0.0005, clipping_epsilon: float = 0.2, num_envs: int = 1024, render_interval: int = 50, batch_size: int = 256, num_minibatches: int = 4, seed: int = 0)

Bases: object

Data class for storing PPO hyperparameters.

alpha: float = 0.1

batch_size: int = 256

clipping_epsilon: float = 0.2

decay_steps: int = 50000000

discounting: float = 0.97

entropy_cost: float = 0.0005

episode_length: int = 1000

learning_rate: float = 0.0001

num_envs: int = 1024

num_evals: int = 1000

num_minibatches: int = 4

num_timesteps: int = 100000000

num_updates_per_batch: int = 4

policy_hidden_layer_sizes: Tuple[int, ...] = (512, 256, 128)

render_interval: int = 50

seed: int = 0

unroll_length: int = 20

value_hidden_layer_sizes: Tuple[int, ...] = (512, 256, 128)

toddlerbot.locomotion.train_mjx module

toddlerbot.locomotion.walk_env module

Module contents