extend-robotics/pick-and-place-foam-block, measured from the public release on 13 September 2026. frame level, every row read
Nobody opens a robot dataset and watches all of it. We read every frame of every episode and report what is in there.
The last fifteen seconds of one episode. The commanded action stops changing and then holds, for ten frames or more, in 97% of the episodes in this release. The longest run is 5.17 seconds of a robot being told to do nothing.
episode 0 of extend-robotics/pick-and-place-foam-block, cut from the release's own video, 15 seconds. Nothing is re-rendered or sped up.
We read every frame of extend-robotics/pick-and-place-foam-block, 30 episodes and 13,500 rows at 30 fps. 97% of episodes end with the commanded action no longer changing for ten frames or more, 97% for at least one frame, and the longest run is 5.17 seconds. We measured the same property on eight public corpora and the rate tracks how the data was produced rather than the task, so it separates scripted takes from human ones.
| property | value |
|---|---|
| episodes | 30 |
| frames | 13,500 |
| frame rate | 30 fps |
| robot | stationary |
| action dimensions | 7 |
| cameras | 2 |
| distinct task strings | |
| shortest episode | not declared |
| median episode | not declared |
| longest episode | not declared |
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 , 0.03 seconds | 29 | 96.7% |
| 5 frames , 0.17 seconds | 29 | 96.7% |
| 10 frames , 0.33 seconds | 29 | 96.7% |
| 30 frames , 1.0 seconds | 29 | 96.7% |
| 60 frames , 2.0 seconds | 29 | 96.7% |
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.