RoboTwin DP Final Checkpoints (archive)
Private archive of final-epoch Diffusion Policy checkpoints for 8 RoboTwin
bimanual tasks. These were exploration-phase DP weights; the DP route was not
closed-loop evaluated (run_eval.py rejects DP execution). Archived only so a
future DP revival can resume from the final epoch without retraining from scratch.
What is stored: one checkpoint per task — the last training epoch only
(550.ckpt/600.ckpt/400.ckpt). Intermediate snapshots (every 50 epochs) were
discarded; they carry no independent value.
Study status: incomplete / DP route closed. This is a data archive, not a released artifact.
| task | file | epoch |
|---|---|---|
| adjust_bottle (joint-5) | adjust_bottle-joint-5/400.ckpt |
400 |
| adjust_bottle (joint-0) | adjust_bottle-joint-0/600.ckpt |
600 |
| beat_block_hammer | beat_block_hammer/600.ckpt |
600 |
| blocks_ranking_rgb | blocks_ranking_rgb/550.ckpt |
550 |
| click_alarmclock | click_alarmclock/600.ckpt |
600 |
| dump_bin_bigbin | dump_bin_bigbin/600.ckpt |
600 |
| grab_roller | grab_roller/600.ckpt |
600 |
| handover_block | handover_block/600.ckpt |
600 |
SHA256 of every checkpoint is in SHA256SUMS.