N0-TWAM post-training — UniVTAC 8 single-arm tasks / absEE (absolute end-effector)
Multi-task post-trained N0-TWAM checkpoint (final release). Trained on marker-less rgb GelSight tactile streams; each task uses its own normalization statistics.
Serving note. Multi-task checkpoints must be served with the per-task
norm / camera+tactile keys / action channels of the task being evaluated — the
pooled envelope in train_meta.json is an unreachable fallback and would
de-normalize actions at the wrong scale. Use the multitask_server config
(TWAM_SERVE_POOL / TWAM_SERVE_TASK env — see
DEPLOY.md);
prompts verbatim from the table below.
Contents: transformer/ (config.json + safetensors, local_tactile tensors = 18)
and train_meta.json (training snapshot).
Task prompts
Send the training prompt verbatim at serve/eval time (also provided as
norm/PROMPTS.json):
| Task | Prompt |
|---|---|
insert_tube |
Insert a tube into a tilted fixture |
insert_hole |
Precision peg-in-hole insertion |
insert_HDMI |
Insert an HDMI connector into a port |
grasp_classify |
Grasp an object and classify it by tactile texture |
pull_out_key |
Untwist and extract a key from a lock |
lift_can |
Rotate a lying can so it stands upright |
lift_bottle |
Grasp and lift a bottle off a surface near a wall |
put_bottle_in_shelf |
Reorient a bottle upright and place it on a shelf |
- Downloads last month
- -