LightwheelAI/Lightwheel-Tasks-G1-WBC, measured from the public release on
16 September 2026.
Every row is read out of the release. Nothing here is estimated and nothing is rounded up.
| episodes declared in the release | 1,600 |
| episodes read | 692 |
| frames read | 1,474,930 |
| data files read | 4 of 9, sampled |
| dimensions in the action space | 23 |
| dimensions constant across every frame read | 2 |
| the two dimensions | base_height_cmd and torso_orientation_roll_cmd |
| range of base_height_cmd over all frames read | 0.0 |
| range of torso_orientation_roll_cmd over all frames read | 0.0 |
| episodes in which base_height_cmd is constant | 692 of 692 |
| declared frame rate | 50 |
The other twenty one dimensions all vary. torso_orientation_pitch and torso_orientation_yaw both move, so it is the roll axis specifically, alongside base height.
Base height and torso roll are the two degrees of freedom that distinguish whole body control from a fixed base arm, and they are the reason to publish a whole body control release at all. A policy trained here sees a constant for both, so it learns to hold them still, and the release teaches nothing about the part of the robot it is named for.
Read any data file and take the range of the base_height_cmd column across all rows. No credentials.
This page is what the release tells you on its own. A full grading removes the part that scores worst, retrains on what is left, and reports whether the smaller corpus scores higher, with rollout counts and intervals. The one result measured on physical hardware is Stera-10M, two arms on a four arm SO-ARM101 bench, 21.9% against 36.9% task success across 1,270 rollouts. The Stera-10M report.