Search papers, labs, and topics across Lattice.
This paper introduces Potential-Guided Policy Optimization (PGPO), a novel approach to enhance credit assignment in multi-turn agentic tasks by leveraging empirical state potentials derived from anchor-state-group return statistics. By enabling cross-trajectory credit propagation, PGPO allows for more nuanced differentiation of actions, particularly in failed trajectories, compared to existing group-based reinforcement learning methods. Experimental results on ALFWorld and WebShop demonstrate that PGPO significantly improves performance and provides more informative failure-side credit signals without incurring additional training costs.
PGPO transforms how we assign credit in multi-turn tasks, allowing effective actions to shine even in the face of failures.
Group-based reinforcement learning (RL) has become an effective paradigm for LLM post-training, but in multi-turn agentic tasks with sparse terminal rewards, it often provides coarse credit for intermediate actions. To obtain more fine-grained credit assignment, recent work such as GiGPO introduces step-level advantages for intermediate actions. However, these step-level signals still rely on the final outcome of each individual trajectory. As a result, actions within failed trajectories can remain poorly differentiated, so effective actions can receive the same unfavorable credit as erroneous ones. In this work, we propose Potential-Guided Policy Optimization (PGPO) for multi-turn agentic tasks. PGPO estimates empirical state potentials from anchor-state-group return statistics within each rollout group. It then derives action advantages from potential differences between adjacent states, enabling cross-trajectory credit propagation. This provides finer-grained step-level credit assignment, especially within failed trajectories. Experiments on ALFWorld and WebShop show strong overall performance relative to recent group-based RL methods. Further analysis provides evidence that PGPO yields more informative failure-side credit signals with negligible training overhead.
