Language

unilab.tasks.motion_tracking.common.manager_terms.MotionCommand¶

class unilab.tasks.motion_tracking.common.manager_terms.MotionCommand[source]¶

Bases: CommandTerm

Device-resident motion reference command for Manager owners.

Parameters:

cfg (MotionCommandCfg)
env (ManagerBasedRlEnv)

Methods

`__init__`(cfg, env)
`bind_read_phase`()	Bind command state that requires the scene's tensor read phase.
`clear_episode_metrics`(reset_mask)	Clear command metrics without a host metric reduction.
`compute`(dt)	Advance the command state by dt.
`post_compute`()	Refresh state that depends on committed command-side simulation writes.
`post_compute_reset`()	Run full-width post-compute after reset.
`refresh_reset`(reset_mask)	Consume the full-width reset mask without row comparison.
`reset`([env_ids, publish_metrics])	Selected-row motion command reset.
`sampler_reset_diagnostics`()

Attributes

`anchor_ang_vel_w`
`anchor_lin_vel_w`
`anchor_pos_w`
`anchor_quat_w`
`body_ang_vel_w`
`body_lin_vel_w`
`body_pos_w`
`body_quat_w`
`command`
`joint_pos`
`joint_vel`
`last_reset_payload_validated`	Whether the current reset payload already passed owner validation.
`name`
`num_envs`
`robot_anchor_ang_vel_w`
`robot_anchor_lin_vel_w`
`robot_anchor_pos_w`
`robot_anchor_quat_w`
`robot_body_ang_vel_w`
`robot_body_lin_vel_w`
`robot_body_pos_w`
`robot_body_quat_w`
`robot_joint_pos`
`robot_joint_vel`
`tensor_body_names`
`tensor_carrier`
`tensor_current_clip_end_frames`
`cfg`

cfg: MotionCommandCfg¶

__init__(cfg, env)[source]¶

Parameters:

cfg (MotionCommandCfg)
env (ManagerBasedRlEnv)

last_step_timing_ms: dict[str, float]¶

last_post_compute_timing_ms: dict[str, float]¶

last_reset_timing_ms: dict[str, float]¶

bind_read_phase()[source]¶

Bind command state that requires the scene’s tensor read phase.

The default command owns no scene reads. Device-resident terms may override this hook to obtain stable public views after the Manager has compiled its packed read plan, avoiding construction-order coupling to EntityScene.read_plan.

Return type:: None

property tensor_carrier: bool¶

property tensor_body_names: tuple[str, ...]¶

property tensor_current_clip_end_frames: Tensor¶

property robot_joint_pos: Tensor¶

property robot_joint_vel: Tensor¶

property robot_body_pos_w: Tensor¶

property robot_body_quat_w: Tensor¶

property robot_body_lin_vel_w: Tensor¶

property robot_body_ang_vel_w: Tensor¶

property robot_anchor_pos_w: Tensor¶

property robot_anchor_quat_w: Tensor¶

property robot_anchor_lin_vel_w: Tensor¶

property robot_anchor_ang_vel_w: Tensor¶

property command: Tensor¶

property joint_pos: Tensor¶

property joint_vel: Tensor¶

property body_pos_w: Tensor¶

property body_quat_w: Tensor¶

property body_lin_vel_w: Tensor¶

property body_ang_vel_w: Tensor¶

property anchor_pos_w: Tensor¶

property anchor_quat_w: Tensor¶

property anchor_lin_vel_w: Tensor¶

property anchor_ang_vel_w: Tensor¶

reset(env_ids=None, *, publish_metrics=True)[source]¶

Selected-row motion command reset.

Parameters:

env_ids (Tensor | slice | None)
publish_metrics (bool)

Return type:

dict[str, float]

sampler_reset_diagnostics()[source]¶

Return type:: dict[str, float]

property last_reset_payload_validated: bool¶: Whether the current reset payload already passed owner validation.

refresh_reset(reset_mask)[source]¶

Consume the full-width reset mask without row comparison.

Parameters:: reset_mask (Tensor)
Return type:: None

post_compute_reset()[source]¶

Run full-width post-compute after reset.

Return type:: None

post_compute()[source]¶

Refresh state that depends on committed command-side simulation writes.

Return type:: None

__call__(*args, **kwargs)¶

Returns the value of the term required by the manager.

Return type:: Any

clear_episode_metrics(reset_mask)¶

Clear command metrics without a host metric reduction.

Parameters:: reset_mask (Tensor)
Return type:: None

compute(dt)¶

Advance the command state by dt.

With env_ids=None (the per-step path) all envs are updated; with env_ids (the reset path) timers and the command update are scoped to those envs. Metrics are refreshed each call; terms may scope per-row metric work to env_ids since other rows are unchanged since the per-step update, or defer row-wise metric work to reset() when reset is the only consumer (e.g. MotionCommand, issue #1355).

dt may be a scalar (all envs) or a per-env tensor (auto-reset path, where freshly reset envs get zero to keep their timers full). A tensor dt requires env_ids=None.

Full-command and metric finite diagnostics run only on the per-step full-batch path. The reset refresh is row-scoped; the immediately preceding command reset or reset-owner transaction remains the authoritative validation boundary for freshly reset rows.

Parameters:: dt (float | Tensor)
Return type:: None

property name: str¶

property num_envs: int¶

metrics: dict[str, Tensor]¶