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_sensorsBuilder:
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.yamlfor IMU sensor names after training.