Independent team details ninth-place 2026 BEHAVIOR policy
First merged summary.
One π0.5-based checkpoint scores 0.392 Q on all 100 tasks in the 2026 BEHAVIOR Challenge.
An independent team, the only independent team in the top 10, describes a ninth-place entry for the 2026 BEHAVIOR Challenge of 100 long-horizon household tasks in OmniGibson. One vision-language-action checkpoint, built on their 2025 π0.5-based winner, covers all 100 tasks. On 50 tasks shared with 2025, Q rises from 0.260 to 0.371 after a 200,000-step run, about one pass over 211 million demonstration frames; all-task Q is 0.392. Fourteen tasks routed to a GPT-6 Astra coding agent average 0.639 Q versus 0.108 for the learned policy. The public submission included only part of those results and scored 0.446 Q with 27.3% full-task success.
- Ninth place and the only independent team in the top 10.
- On 50 shared 2025 tasks, Q rises from 0.260 to 0.371.
- A 200k-step run is about one pass over 211 million frames.
- GPT-6 Astra averages 0.639 Q on 14 tasks versus 0.108.
- Public submission scores 0.446 Q with 27.3% full-task success.
- Six simulators on one GPU run faster than real time.
What each outlet adds
beyond the shared facts- arXivNinth-place BEHAVIOR scores, Astra routing, and training scale.
- Products
- OmniGibson
- AI models
- π0.5GPT-6 Astra
Coverage
oldest first · one row per article · UTC- · 2d ago
arXiv cs.AI / cs.LG / cs.CLFirst report
Off-policy RL entry scores 0.990 RDS on CoRL 2026 RoboEval
A π0.5 policy trained with AWR, RECAP, generated demos and failure recoveries also scores 0.987 on official seeds.
- · 11h ago
One VLA policy places 9th in the 2026 BEHAVIOR Challenge
A single π0.5-based checkpoint scores 0.392 Q on 100 tasks; GPT-6 Astra averages 0.639 Q on 14 routed tasks.