BEHAVIOR Challenge 2026 β€” task 29 (clean_up_your_desk), low-lr fine-tune (skill stages), step 1999

Selected checkpoint for the BEHAVIOR-1K 2026 submission.

Model

  • Architecture: the 2025 BEHAVIOR Challenge winner's pi0.5-based policy (IliaLarchenko/behavior-1k-solution, ~3B params, task-ID embeddings, no language prompt).
  • Init: the winner's task-specific checkpoint_1, fine-tuned (not used unchanged).
  • Training config: pi_behavior_b1k_2026_task29_lowlr (in our pipeline repo's behavior-1k-solution/src/b1k/training/config.py), with skill-annotation subtask stages.
  • Data: 2026 challenge demos, task 29 episodes 5800-5979 (180 demos), RGB cameras only.
  • Recipe: 2000 steps from the winner's checkpoint_1, batch 16, cosine lr 5e-6 -> 1e-6 (warmup 100), skill stages. This is the final step 1999 (no screening; straight to the official run).
  • Checkpoint is params-only (no optimizer state): it can be served or used as an init, not --resumed.

Layout

  • params/ β€” orbax/ocdbt parameters
  • assets/ β€” normalization stats
  • _CHECKPOINT_METADATA

Serve with scripts/serve_b1k.py --task-id 29 policy:checkpoint --policy.config pi_behavior_b1k_2026_task29_lowlr --policy.dir <this dir>.

Evaluation (q = fraction of goal predicates newly satisfied)

Scored the 2026 leaderboard way: public instances 0-9 (ids 301-310), one rollout each, default 1.5x timeouts, task score = sum / 10.

Leaderboard score: q = 0.1000 (sum 1.0000 over 301-310; full completion: none).

instance 301 302 303 304 305 306 307 308 309 310
q 0 0 .182 0 0 .091 .091 .364 .273 0
Downloads last month

-

Downloads are not tracked for this model. How to track
Video Preview
loading