unilab.tasks.locomotion.common.sensor_reward_terms¶
Sensor-driven reward terms shared by Manager-Based locomotion families.
Biped and quadruped locomotion families port legacy reward equations that read
IMU-style XML sensors by name. This module owns those equations once, on top
of the SensorTermBase cold-path binding contract from manager_terms.py:
the robot’s sensor name is a cfg parameter (params={"sensor_name": ...}),
so each family binds its own sensors from its owner YAML instead of
subclassing. Plain action/state penalties and the shared term bases live in
manager_terms.py.
Classes
Penalty for roll/pitch angular velocity. |
|
Penalty for vertical (z) linear velocity. |
|
Penalty for deviation from upright (horizontal upvector components). |
|
Exponential reward for tracking commanded yaw angular velocity. |
|
Exponential reward for tracking commanded xy linear velocity. |
- class unilab.tasks.locomotion.common.sensor_reward_terms.ang_vel_xy[source]¶
Bases:
_Vec3SensorTermPenalty for roll/pitch angular velocity.
- Parameters:
cfg (
ManagerTermBaseCfg)env (
ManagerBasedRlEnv)
- class unilab.tasks.locomotion.common.sensor_reward_terms.lin_vel_z[source]¶
Bases:
_Vec3SensorTermPenalty for vertical (z) linear velocity.
- Parameters:
cfg (
ManagerTermBaseCfg)env (
ManagerBasedRlEnv)
- class unilab.tasks.locomotion.common.sensor_reward_terms.orientation[source]¶
Bases:
_Vec3SensorTermPenalty for deviation from upright (horizontal upvector components).
- Parameters:
cfg (
ManagerTermBaseCfg)env (
ManagerBasedRlEnv)