Same libero_spatial task 0 demo-0 init state. 1 episode, max_steps=600,
--teleport --zero_rotation. All three models hit max_steps (0% SR).
| Model | val_px_err | SR | pred jerk (px) | Video |
|---|---|---|---|---|
| (A) 2D AR | 8.0 (3 ep) | 0% | 1.89 | ▶ play A |
| (B) Voxel + abs xyz | 12.4 (2 ep pilot) | 0% | 0.25 | ▶ play B |
| (C) Voxel + EEF-rel xyz | 13.3 (2 ep pilot) | 0% | 0.06 | ▶ play C |
White dot = current EEF projected to image. Green crosshair = predicted next-EEF cell. Label shows step idx and gripper sign.
All 3 confirm the under-training pathology: predicted target stays in the current 8-px grid cell, action deltas are 1-2mm/step, nothing happens before max_steps. The voxel models are MORE frozen (jerk 7–30× lower) than the 2D AR — consistent with them being 2 epochs vs 3 epochs and (for C) hitting the rel-PE unit-mismatch bug I flagged.