YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
AIC rosbags (AI for Industry Challenge: cable insertion)
Backup of the 660 rosbag2 recordings from our local self-scoring and data-collection runs for Intrinsic's
AI for Industry Challenge (UR5e cable insertion, Gazebo aic_eval image sha256:9aa2ffdb…, built 2026-04-16).
The engine container recorded them automatically, one per trial, between 2026-07-02 and 2026-07-11.
Contents: robot transforms, the scorer's ground-truth transforms, joint states, controller state, wrist force/torque and
the policy's commanded poses. No camera images. Full description: ROSBAGS.md.
Layout
Paths mirror the original disk layout. Each .mcap is zstd-compressed (.mcap.zst, level 3); metadata.yaml is unchanged.
| Path | What |
|---|---|
aic_results/<run>/bag_<trial>_<UTC time>/ |
scored eval runs (score.sh); scoring.yaml and policy.log sit next to the bags |
aic_results/tuner/chunk_0/ |
InsertTuner sweep (70 bags) |
aic_results_archive/collect/chunk_<k>/ |
CheatCode data collection for the perception_v1 dataset |
aic_results_archive/parcollect/chunk_<k>/ |
CheatCode data collection for perception_native (pass 1, before 18:00 UTC, chunks 0–4) and perception_v2 (pass 2, chunks 0–17). These bags are the only copy of the expert actions (/aic_controller/pose_commands) for those two datasets. |
rosbag_inventory.csv |
one row per run: bags, size, time span, policy, score |
MANIFEST.jsonl |
one row per bag: original size and sha256, compressed size and sha256 |
SHA256SUMS |
sha256 of every original .mcap, in sha256sum -c format |
Restore
hf download sangramrout/air_data --include "aic_results/seat_v4_4/*" --local-dir aic_bags # or no --include for everything
cd aic_bags
find . -name '*.mcap.zst' -exec zstd -d --rm {} \;
sha256sum -c --ignore-missing SHA256SUMS
ros2 bag info aic_results/seat_v4_4/bag_trial_1_* # readable by rosbag2 (MCAP storage plugin)
Re-running a policy with the same config records a new bag; it does not reproduce these. Learned policies vary a lot
from run to run. See "Reproducibility" in ROSBAGS.md.