HapticWAM student
HapticWAM: Distilling Imagined Touch into a World-Action Model without Inference-Time Tactile Sensing β paper arXiv:2609.23888, submitted to ICRA 2027. Code: github.com/Advanced-Robotic-Manipulation/HapticWAM Β· all repos: HapticWAM β ICRA 2027.
The distilled student: the model that runs on the rig. It is distilled from the HapticWAM teacher and takes pad-free inputs, so it does not need the fingertip tactile streams at inference time.
The checkpoint the paper deploys is hid_simft/student_001000.pt.
Folders
| Folder | Files | Size | What it is |
|---|---|---|---|
hid_simft/ |
6 | 1.49 GB | the deployed student. Steps 250 / 500 / 750 / 1000, plus the distillation logs distill_A.log and extend_A.log. Distilled from teacher_v6_simft/teacher_002000.pt. |
hid_ftA_r2_nowrist/ |
5 | 1.49 GB | the wrist-masked (sensor-free) student arm, steps 500β2000, log stage4_nowrist.log. Distilled from the ftA teacher. |
hid_ftA_r2_nowrist_cont/ |
6 | 1.49 GB | the continuation of that arm, steps 2500β4000, logs cont_nowrist_b.log and stage5.log. |
eval/ |
24 | β | evaluations of exactly the checkpoints above: hid_simft_student_* at each rung, and nowrist_student_* / nowrist_cont_student_* at each rung. JSON plus log for each. |
index.jsonl lists every file here with its size, LFS sha256 and the path it was copied from.
Which arm is which
hid_simft is the deployed arm and the one every headline number refers to. The two
nowrist folders are the sensor-free ablation: the same distillation recipe with the wrist
tactile input masked out. They were evaluated offline and never run on the rig.
Teacher and data
Teacher: armteam/hapticwam-teacher.
Data: armteam/hapticwam-teleop-dataset,
1,115 episodes, manifests_v6.tar, plus the rig rollouts in
armteam/hapticwam-rollouts.
The eval files here are the subset that scores the checkpoints in this repo. The complete
sweeps, including every arm that is not in this repo, live in armteam/hapticwam-ablations.
Part of the HapticWAM release
Ten repos on the hub, gathered in the HapticWAM β ICRA 2027 collection.
| Repo | Kind | Holds |
|---|---|---|
armteam/hapticwam-teacher |
model | the tactile-input teacher. Deployed checkpoint teacher_v6_simft/teacher_002000.pt; also holds the Cosmos prompt cache text_embeddings.pt |
armteam/hapticwam-student β you are here |
model | the distilled pad-free student, the model that runs on the rig. Deployed checkpoint hid_simft/student_001000.pt |
armteam/hapticwam-baselines |
model | the pi0.5, Diffusion Policy and X-VLA baselines at the deployed steps |
armteam/hapticwam-ablations |
model | every training arm that is not deployed, and the complete evaluation sweeps |
armteam/hapticwam-teleop-dataset |
dataset | the training corpus β 1,115 teleoperated episodes, packed per task |
armteam/hapticwam-teleop-raw |
dataset | the same teleoperation as loose, as-recorded sessions (provenance) |
armteam/hapticwam-sim-episodes |
dataset | Isaac Sim expert episodes, used for the sim fine-tune |
armteam/hapticwam-rig-episodes |
dataset | the closed-loop rig takes the reported numbers are computed from |
armteam/hapticwam-rollouts |
dataset | policy-driven rollouts β the DAgger rounds and the deploy days |
armteam/hapticwam-evidence |
dataset | per-take evidence behind the paper's tables β scored CSVs, probe JSONs, figures |
Code, training and deployment scripts: github.com/Advanced-Robotic-Manipulation/HapticWAM.
Licence
Model weights and the HapticWAM code: Apache-2.0. The datasets listed above: CC-BY-4.0.