Language

unilab.tasks.manipulation.stewart.balance¶

Manager-Based terms for Stewart-platform ball balancing.

Hydra owns the production task declaration. This module contains only the task-specific NumPy terms and the generic Manager-Based registry binding.

Functions

`center_reward`(env, state_term_name)
`fall_reward`(env, state_term_name)
`progress_reward`(env, state_term_name)
`still_reward`(env, state_term_name)

Classes

`StewartBalanceState`	Termination-owned progress, stillness, success, and fall state.
`StewartBallReset`	Sample the ball uniformly within a disk via root-state entity writes.
`StewartObservation`	Legacy 15-D observation with per-environment filtered finite differences.
`StewartTiltAction`	Vectorized tilt IK using only the public entity state/control facade.
`StewartTiltActionCfg`	Two-axis tilt action converted to six Stewart actuator targets.

class unilab.tasks.manipulation.stewart.balance.StewartBalanceState[source]¶

Bases: ManagerTermBase

Termination-owned progress, stillness, success, and fall state.

Parameters:

cfg (ManagerTermBaseCfg)
env (_StewartEnv)

__init__(cfg, env)[source]¶

Parameters:

cfg (ManagerTermBaseCfg)
env (_StewartEnv)

property last_counter: int¶

reset(env_ids=None)[source]¶

Resets the manager term.

Parameters:: env_ids (ndarray | slice | None)
Return type:: None

class unilab.tasks.manipulation.stewart.balance.StewartBallReset[source]¶

Bases: ManagerTermBase

Sample the ball uniformly within a disk via root-state entity writes.

Parameters:

cfg (ManagerTermBaseCfg)
env (ManagerBasedRlEnv)

__init__(cfg, env)[source]¶

Parameters:

cfg (ManagerTermBaseCfg)
env (ManagerBasedRlEnv)

class unilab.tasks.manipulation.stewart.balance.StewartObservation[source]¶

Bases: ManagerTermBase

Legacy 15-D observation with per-environment filtered finite differences.

Parameters:

cfg (ManagerTermBaseCfg)
env (_StewartEnv)

__init__(cfg, env)[source]¶

Parameters:

cfg (ManagerTermBaseCfg)
env (_StewartEnv)

property relative_xy: ndarray¶

property velocity_xy: ndarray¶

property ball_pos: ndarray¶

reset(env_ids=None)[source]¶

Resets the manager term.

Parameters:: env_ids (ndarray | slice | None)
Return type:: None

snapshot(env)[source]¶

Parameters:: env (_StewartEnv)
Return type:: ndarray

class unilab.tasks.manipulation.stewart.balance.StewartTiltAction[source]¶

Bases: ActionTerm

Vectorized tilt IK using only the public entity state/control facade.

Parameters:

cfg (StewartTiltActionCfg)
env (ManagerBasedRlEnv)

cfg: StewartTiltActionCfg¶

__init__(cfg, env)[source]¶

Parameters:

cfg (StewartTiltActionCfg)
env (ManagerBasedRlEnv)

property action_dim: int¶

property raw_action: ndarray¶

property executed_action: ndarray¶

property target_tilt_deg: ndarray¶

property neutral_leg_lengths: ndarray¶

leg_control_for_tilt(target_tilt_rad)[source]¶

Return six actuator controls for (roll, pitch) radians.

Parameters:: target_tilt_rad (ndarray)
Return type:: ndarray

process_actions(actions)[source]¶

Parameters:: actions (ndarray)
Return type:: None

apply_actions()[source]¶

Return type:: None

reset(env_ids=None)[source]¶

Resets the manager term.

Parameters:: env_ids (ndarray | slice | None)
Return type:: None

class unilab.tasks.manipulation.stewart.balance.StewartTiltActionCfg[source]¶

Bases: ActionTermCfg

Two-axis tilt action converted to six Stewart actuator targets.

Parameters:

entity_name (str)
clip (dict[str, tuple] | None)
actuator_names (tuple[str, ...] | list[str])
top_body_name (str)
ball_body_name (str)
leg_body_names (tuple[str, ...] | list[str])
top_connect_body_names (tuple[str, ...] | list[str])
raw_action_clip (tuple[float, float] | list[float])
target_rotation_limit_deg (float)
action_smooth (float)
center_control_radius (float)
center_control_min_gain (float)

actuator_names: tuple[str, ...] | list[str]¶

top_body_name: str¶

ball_body_name: str¶

leg_body_names: tuple[str, ...] | list[str]¶

top_connect_body_names: tuple[str, ...] | list[str]¶

raw_action_clip: tuple[float, float] | list[float]¶

target_rotation_limit_deg: float¶

action_smooth: float¶

center_control_radius: float¶

center_control_min_gain: float¶

build(env)[source]¶

Build the action term from this config.

Parameters:: env (ManagerBasedRlEnv)
Return type:: StewartTiltAction

__init__(*, entity_name, clip=None, actuator_names, top_body_name, ball_body_name, leg_body_names, top_connect_body_names, raw_action_clip, target_rotation_limit_deg, action_smooth, center_control_radius, center_control_min_gain)¶

Parameters:

entity_name (str)
clip (dict[str, tuple] | None)
actuator_names (tuple[str, ...] | list[str])
top_body_name (str)
ball_body_name (str)
leg_body_names (tuple[str, ...] | list[str])
top_connect_body_names (tuple[str, ...] | list[str])
raw_action_clip (tuple[float, float] | list[float])
target_rotation_limit_deg (float)
action_smooth (float)
center_control_radius (float)
center_control_min_gain (float)

entity_name: str¶: Name of the entity in the scene that this action term controls.

unilab.tasks.manipulation.stewart.balance.center_reward(env, state_term_name)[source]¶

Parameters:

env (_StewartEnv)
state_term_name (str)

Return type:

ndarray

unilab.tasks.manipulation.stewart.balance.fall_reward(env, state_term_name)[source]¶

Parameters:

env (_StewartEnv)
state_term_name (str)

Return type:

ndarray

unilab.tasks.manipulation.stewart.balance.progress_reward(env, state_term_name)[source]¶

Parameters:

env (_StewartEnv)
state_term_name (str)

Return type:

ndarray

unilab.tasks.manipulation.stewart.balance.still_reward(env, state_term_name)[source]¶

Parameters:

env (_StewartEnv)
state_term_name (str)

Return type:

ndarray