act_flip_v2_base
v2 expert as is, 100 demos, front camera, dense checkpoints
Jacob's from-scratch ACT (scripts/act.py, scripts/train_act.py in https://github.com/JaHeRoth/robot-learning). Checkpoints uploaded: best = step 10000, final = step 30000. Raw torch.save dicts with model_state, opt_state, stats.
| step | success | 95% CI |
|---|---|---|
| 5000 | 0.734 | [0.694, 0.771] |
| 10000 | 0.966 | [0.946, 0.979] |
| 30000 | 0.944 | [0.920, 0.961] |
500 held-out seeds (800000-800499), n_action_steps 40, horizon 450