Towards Efficient Online Tuning of VLM Agents via Counterfactual Soft Reinforcement Learning

TL;DR

This paper introduces CoSo, a novel counterfactual soft reinforcement learning method that improves online fine-tuning of vision-language model agents by focusing exploration on critical tokens, leading to better efficiency and performance.

Contribution

The paper proposes CoSo, a new RL approach that uses counterfactual reasoning to target important tokens in textual actions, enhancing online exploration for VLM agents.

Findings

CoSo improves exploration efficiency in diverse tasks.

It achieves consistent performance gains over prior methods.

Theoretical guarantees support CoSo's convergence and policy improvement.

Abstract

Online fine-tuning vision-language model (VLM) agents with reinforcement learning (RL) has shown promise for equipping agents with multi-step, goal-oriented capabilities in dynamic environments. However, their open-ended textual action space and non-end-to-end nature of action generation present significant challenges to effective online exploration in RL, e.g., explosion of the exploration space. We propose a novel online fine-tuning method, Counterfactual Soft Reinforcement Learning (CoSo), better suited to the textual output space of VLM agents. Compared to prior methods that assign uniform uncertainty to all tokens, CoSo leverages counterfactual reasoning to dynamically assess the causal influence of individual tokens on post-processed actions. By prioritizing the exploration of action-critical tokens while reducing the impact of semantically redundant or low-impact tokens, CoSo enables a more targeted and efficient online rollout process. We provide theoretical analysis proving CoSo's convergence and policy improvement guarantees, and extensive empirical evaluations supporting CoSo's effectiveness. Our results across a diverse set of agent tasks, including Android device control, card gaming, and embodied AI, highlight its remarkable ability to enhance exploration efficiency and deliver consistent performance gains. The code is available at https://github.com/langfengQ/CoSo.