Towards Practical World Model-based Reinforcement Learning for Vision-Language-Action Models
This addresses the practical deployment challenge for robotic control using VLA models, though it appears incremental as it builds on existing multimodal models and model-based RL techniques.
The paper tackles the problem of finetuning Vision-Language-Action models with reinforcement learning by avoiding costly real-world interactions through world model training, and demonstrates significant improvements in policy performance and sample efficiency in simulation and real-world tasks.
Vision-Language-Action (VLA) models show strong generalization for robotic control, but finetuning them with reinforcement learning (RL) is constrained by the high cost and safety risks of real-world interaction. Training VLA models in interactive world models avoids these issues but introduces several challenges, including pixel-level world modeling, multi-view consistency, and compounding errors under sparse rewards. Building on recent advances across large multimodal models and model-based RL, we propose VLA-MBPO, a practical framework to tackle these problems in VLA finetuning. Our approach has three key design choices: (i) adapting unified multimodal models (UMMs) for data-efficient world modeling; (ii) an interleaved view decoding mechanism to enforce multi-view consistency; and (iii) chunk-level branched rollout to mitigate error compounding. Theoretical analysis and experiments across simulation and real-world tasks demonstrate that VLA-MBPO significantly improves policy performance and sample efficiency, underscoring its robustness and scalability for real-world robotic deployment.