State-estimation observations

State-estimation observations#

Task id: Wbc-G1-SE · In-tree reference on robot g1.

Default WBC stack + state-estimation actor observations — anchor pose tracking error and base linear velocity on top of Default WBC stack.

  • Presets: apply_wbc + apply_se_actor + wire_g1_imu_sensors

  • Builder: g1_wbc_se_env_cfg()

What this task adds#

On top of Default WBC stack, apply_se_actor (presets/se_actor.py):

Removes from actor: ref_base_height, ref_gravity_b, projected_gravity

Adds: motion_anchor_pos_error_w, motion_anchor_ori_error, base_lin_vel (IMU wiring in the robot entity builder)

Dim rules: Observations.

Train & play#

uv run wbc-mjlab-train --task Wbc-G1-SE --dataset samples
uv run wbc-mjlab-play --task Wbc-G1-SE --dataset samples --viewer viser

Logs: logs/rsl_rl/wbc_g1_se/<run>/

Tips#

  • For deploy export, prefer Default WBC stack unless your runtime implements the same SE observation terms.

  • Check env.yaml for IMU sensor names after training.