ExToken: Structured Exploration for Efficient Vision-Language-Action Reinforcement Fine-tuning

2026-07-14

AI Summary

This work addresses the challenges of inefficient exploration and underutilized samples in vision-language-action (VLA) reinforcement learning, which stem from high interaction costs. To tackle these issues, the authors propose ExToken, a framework that leverages offline demonstrations to construct a discrete behavioral prior represented as exploration tokens. A state-conditional token selector is introduced to enable structured and state-adaptive exploration. ExToken is the first approach to integrate discrete behavioral priors with a dynamic token selection mechanism while maintaining consistency between training and deployment. Experimental results demonstrate that ExToken significantly accelerates convergence and improves task success rates under extremely limited interaction budgets, exhibiting strong performance and robustness in both simulated and real-world robotic manipulation tasks.

Abstract

Reinforcement Learning (RL) has demonstrated significant potential for improving Vision-Language-Action (VLA) models on complex manipulation tasks. However, its practical scalability remains severely limited by the substantial cost of environmental interactions. In this work, we first investigate the exploration stagnation bottleneck in current VLA-RL frameworks and reveal that trajectory diversity is fundamentally more important to sample efficiency than the sheer quantity of collected rollouts. Motivated by these insights, we introduce RL Exploration Token (ExToken), a simple yet general framework that condition VLA policies on discrete behavioral priors derived from offline demonstrations for structured exploration. By conditioning the policy on different tokens during rollout collection, ExToken encourages the agent to explore diverse behavioral modes, substantially improving state-action coverage and exploration efficiency. To bridge exploration during training with deterministic inference at deployment, ExToken further incorporates a state-conditioned token selector that adaptively predicts effective behavioral modes for unseen scenarios. Extensive experiments across simulated and real-world robotic manipulation tasks demonstrate that ExToken consistently accelerates convergence, improves task performance, and exhibits strong robustness under highly constrained interaction budgets.

Problem

• Vision-Language-Action
• Reinforcement Learning
• Exploration Stagnation
• Sample Efficiency
• Robotic Manipulation

Innovation

• structured exploration
• vision-language-action
• reinforcement fine-tuning
• behavioral priors
• sample efficiency