v1.0: simulation checkpoints and dataset

Simulation checkpoints and dataset for Dexterity from Supervised Play (DSP).

file	size	contents
dsp_sim_play_policy.pt	30 MB	DiT diffusion play policy (7.5M params), the paper's sim checkpoint; inference-only (EMA weights, optimizer stripped)
dsp_sim_world_model.pt	157 MB	DiT diffusion world model (39M params, 50-step horizon, 3-frame history), the paper's sim checkpoint
dsp_sim_dataset_v1.tar	232 MB	2 human glove teleop takes (29,474 frames) + 5.0 h autonomous play (1,081,500 frames, 206 episodes); see docs/DATASET.md
SHA256SUMS	—	checksum of the dataset tar

Get them with the scripts in the repository, which verify every checksum:

bash quickstart/download_checkpoints.sh   # -> checkpoints/
bash scripts/download_dataset.sh          # -> data/dsp_sim_dataset/

sha256:

744a4159f7842e6e2c7dbac57748c74c771f5c13ffdfa7b302757b6f75b8b463  dsp_sim_play_policy.pt
3d337874d91b06f12fac94ab62f51a02772ce819191b2002ecb8dc1e713b1498  dsp_sim_world_model.pt
6c43b25588765b9dce571f5b50342f969d132b058742f9cf02e47d7fd3039d37  dsp_sim_dataset_v1.tar