2512.06951

Total: 1

#1 Task adaptation of Vision-Language-Action model: 1st Place Solution for the 2025 BEHAVIOR Challenge [PDF] [Copy] [Kimi1] [REL]

Authors: Ilia Larchenko, Gleb Zarin, Akash Karnatak

We present a vision-action policy that won 1st place in the 2025 BEHAVIOR Challenge - a large-scale benchmark featuring 50 diverse long-horizon household tasks in photo-realistic simulation, requiring bimanual manipulation, navigation, and context-aware decision making. Building on the Pi0.5 architecture, we introduce several innovations. Our primary contribution is correlated noise for flow matching, which improves training efficiency and enables correlation-aware inpainting for smooth action sequences. We also apply learnable mixed-layer attention and System 2 stage tracking for ambiguity resolution. Training employs multi-sample flow matching to reduce variance, while inference uses action compression and challenge-specific correction rules. Our approach achieves 26% q-score across all 50 tasks on both public and private leaderboards.

Subjects: Robotics , Artificial Intelligence , Computer Vision and Pattern Recognition , Machine Learning

Publish: 2025-12-07 18:08:45 UTC