yaak-ai/L2D-v3, measured from the public release on 13 September 2026. frame level, every row read
We read every frame of yaak-ai/L2D-v3, 10,000 episodes and 2,298,363 rows. In 9657 of them the dimension named discrete[0] holds one value for the whole episode. It is worth knowing whether that is intended, because for that share of episodes the policy is learning a constant on that channel.
| property | value |
|---|---|
| episodes | 10,000 |
| frames | 2,298,363 |
| frame rate | 10 fps |
| robot | KIA Niro EV 2023 |
| action dimensions | 5 |
| cameras | 7 |
| distinct task strings | 1 |
| shortest episode | 51 frames |
| median episode | 300 frames |
| longest episode | 600 frames |
A trailing hold is the number of frames at the end during which the commanded action does not change by more than 1e-4 on any component. Seconds come from the frame rate the release declares.
| trailing hold at least | episodes | share |
|---|---|---|
| 1 frames , 0.1 seconds | 2,340 | 23.4% |
| 5 frames , 0.5 seconds | 215 | 2.2% |
| 10 frames , 1.0 seconds | 51 | 0.5% |
| 30 frames , 3.0 seconds | 8 | 0.1% |
| 60 frames , 6.0 seconds | 6 | 0.1% |
The release is a meta directory plus one or more parquet files. Read data/chunk-000/file-000.parquet, group rows by the episode ranges in meta/episodes, and compute the same columns. No credentials.
This page is what a published release will tell you on its own. The full grading removes the part that scores worst and retrains on what is left, then reports whether the smaller corpus scores higher, with rollout counts and intervals. The one result measured on physical hardware is Stera-10M, two arms on a four arm SO-ARM101 bench, 21.9% against 36.9% task success across 1,270 rollouts.