File size: 773 Bytes
0a007bc ff78faf 46d2599 ff78faf 0a007bc ff78faf 0a007bc | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 | # BEHAVIOR 2026 Self-Evaluation
This dataset contains three independent evaluation rounds: `round0/`, `round1/`, and `round2/`. Each round covers 100 tasks with instances 301-310, totaling 1,000 episodes. Results are reported separately for each round.
Files are named `{task}_{instance}_{rollout}`.
| Directory | Successful episodes | Macro Q-score | Video size |
| --- | ---: | ---: | ---: |
| `round0/` | 398 | 0.604701 | 29.50 GB |
| `round1/` | 416 | 0.617207 | 29.32 GB |
| `round2/` | 414 | 0.618579 | 29.06 GB |
Each round contains `metrics/`, `videos/`, `robot/r1pro.yaml`, `wrapper/default_wrapper.py`, and `selection_index.json`.
Macro Q-score is calculated by averaging the 10 instance Q-scores within each task, then averaging the 100 task means equally.
|