ポリシーパッチを超えて:RL専門家を視覚-言語-行動ポリシーへ蒸留する
Beyond Policy Patches: Distilling RL Specialists into Vision-Language-Action Policies
強化学習の専門家ポリシーを価値誘導型蒸留で単一のフローマッチングVLAに統合し、展開時に専門家なしで6つの実機タスク全てで100%成功を達成した。
詳しい要約
1. どんなもの?
2. 先行研究と比べてどこがすごい?
3. 技術・手法の肝は?
4. どうやって有効だと検証した?
5. 議論はある?
6. 次に読むべき論文は?
※ AIが要旨から生成した要約です。正確性は原文をご確認ください。
著者: Zhennan Jiang, Tingying Xie, Shilong Ji, Shu'ang Yu, Tianxing Zhou, Yuhui Chen, Jiahuan Gao, Kaizhi Xie, Yuanfan Xu, Feng Gao, Yu Wang, Haoran Li, Dongbin Zhao, Chao Yu
分類: cs.RO
原文アブストラクト
Vision-language-action (VLA) policies can handle noncritical stages of long-horizon manipulation yet fail during short precision-critical phases. Lightweight reinforcement learning (RL) specialists can address these failures without directly applying RL to the full VLA, but retain auxiliary controllers and switching mechanisms at deployment. We study how to transfer their capabilities into a single flow-matching VLA. Direct imitation of specialist rollouts can reproduce action jitter and leave the VLA unprepared for states encountered during its own execution. We introduce REPAIR, a value-guided distillation framework that combines offline initialization with on-policy refinement. At each critical-phase observation, REPAIR queries the corresponding specialist for multiple action candidates and weights them using its critic to construct a supervision distribution. A shared weighted flow-matching objective trains the VLA first on observations from specialist-assisted rollouts, then on observations collected through its own execution. Specialists provide supervision without taking control during on-policy collection and are removed at deployment. Across six real-robot tasks with 30 evaluation trials per task, REPAIR achieves 100% success on all six tasks, with lower mean end-effector jerk than specialist-assisted execution on every task.