ReflectPO: Reflection-Guided Policy Optimization for Long-Horizon LLM Agents
Rumei Li and Xunliang Cai (Meituan) propose ReflectPO, an RL framework that trains long-horizon tool agents to recover from their own early errors.
Ask this paper
Reflection-guided sampling. Branch differences and repeated visits in trajectories are turned into hints for error recovery during rollout.
Tree credit. Sampled trajectories are organized into trees, and action-level credit comes from parent-child reward differences plus recovery gains.
Result. Outperforms existing baselines in most domains of two long-horizon tool-use benchmarks and has the best average.
Behavior. Analyses show improved recognition of deviations and recovery from erroneous states.
Abstract
With external tools, large language models (LLMs) can tackle complex tasks through sustained environmental interaction. LLM-based agents are therefore increasingly used to tackle complex long-horizon tasks in real-world environments. However, these tasks involve sequential, interdependent decisions and changes in environmental states. Early execution errors affect subsequent interaction histories and environmental states, and their effects accumulate along the decision chain. To address this challenge, we propose Reflection-Guided Policy Optimization (ReflectPO), a reinforcement learning framework for autonomous error recovery in long-horizon tasks. ReflectPO comprises two core components. The first is Reflection-Guided Sampling. We use branch differences and repeated visits in interaction trajectories to construct hints for error recovery. The second is Tree-Structured Credit Assignment. We organize sampled trajectories into trees and assign action-level credit using parent--child reward differences and recovery gains. Experiments over two long-horizon tool-use benchmarks show that ReflectPO outperforms multiple existing baselines in most domains. It also achieves the best overall performance on average. Further analyses show that ReflectPO effectively improves agents' ability to recognize their own deviations, recover from erroneous states, and continue making progress toward task completion.