bf16 rollout capture from the LingBot-VA world-action model on the RoboTwin 2.0 simulator: per-chunk observations, video latents, and predicted actions recorded over 250 episodes.

Provenance


Source model	LingBot-VA (`robbyant/lingbot-va-posttrain-robotwin`, 5.09B bf16 Wan transformer)
Simulator	RoboTwin 2.0
Tasks	50 RoboTwin tasks
Episodes	5 per task → 250 episodes
Scene config	`demo_randomized` (per-episode randomized background / lighting / table height)
Precision	bf16 (unquantized baseline rollouts)
Chunks	2079 obs / latent / action chunks total (~8 per episode)
Size	~12 GB (6237 `.pt` files)

Composition

One directory per episode, named <language_prompt>_<YYYYMMDD_HHMMSS>/. Each directory holds three tensor types per autoregressive chunk index i:

<episode>/
  obs_data_<i>.pt   # raw observation window fed to the model at chunk i (pre-VAE)
  latents_<i>.pt    # VAE-encoded video latents at chunk i
  actions_<i>.pt    # model-predicted action chunk at chunk i

Counts: 2079 files of each type (obs_data / latents / actions), across 250 episodes. All tensors are saved with torch.save; load with torch.load(path, weights_only=False, map_location="cpu"). NumPy arrays load without torch. Image arrays are uint8; latents / actions are bfloat16.

`obs_data_<i>.pt` — `list` of per-frame observation records

Each list element is one frame, a dict in the RoboTwin / aloha-agilex layout:

key	type / shape	meaning
`observation.images.cam_high`	`uint8 [240, 320, 3]`	head camera RGB (HWC)
`observation.images.cam_left_wrist`	`uint8 [240, 320, 3]`	left-wrist camera RGB
`observation.images.cam_right_wrist`	`uint8 [240, 320, 3]`	right-wrist camera RGB
`observation.state`	`float64 [14]`	proprioceptive state (dual 7-DoF arms)
`task`	`str`	language instruction

The list is cumulative — it grows as the rollout proceeds (4 records at chunk 0, 8 at chunk 2, … i.e. the full observation history up to chunk i).

`latents_<i>.pt` — `bfloat16 [1, 48, 2, 24, 20]`

[batch, latent_channels=48, temporal_frames=2, H_lat=24, W_lat=20] — the WAN VAE-encoded video latent for the 2 frames generated at chunk i.

`actions_<i>.pt` — `bfloat16 [1, 30, 2, 16, 1]`

[batch, action_dim=30, temporal_frames=2, steps=16, 1] — the model-predicted action chunk for chunk i (fixed shape per chunk, not cumulative).

Chunk indexing

Chunk indices are even and contiguous (0, 2, 4, 6, …): the WAN VAE temporally downsamples 2:1, so each autoregressive step emits 2 latent frames, and <i> is the frame-start id of that step (hence the step of 2). The set of indices present per episode runs 0 … 2·(num_chunks − 1); episode length varies with task horizon. Episode directories are flattened at the repo root (the upload tool did not preserve a parent prefix); each name is unique.

Download

from huggingface_hub import snapshot_download
snapshot_download(
    repo_id="JingwuLuo/LingBot-VA_RoboTwin_clibration_data",
    repo_type="dataset",
    local_dir="calib_capture",
)

The repo is ~12 GB across 6k+ small files. On a rate-limited (free) HF account you may hit HTTP 429 — just re-run snapshot_download; it resumes and finishes over a few waves.

License

Released under CC-BY-4.0. If you use this data, please credit the LingBot-VA and RoboTwin authors and link back to this dataset.

Downloads last month: 3,270

Total file size:

12.3 GB

LingBot-VA × RoboTwin Calibration Data