Search-P1: Path-Centric Reward Shaping for Stable and Efficient Agentic RAG Training
This work improves the training efficiency and performance of agentic RAG systems, which are crucial for LLMs tackling complex multi-step reasoning tasks, by providing a more informative reward signal.
This paper addresses the limitations of sparse outcome rewards and low sample efficiency in RL-based training for agentic RAG by introducing Search-P1. This framework uses path-centric reward shaping, which evaluates reasoning trajectories structurally and extracts learning signals from failed samples, achieving an average accuracy gain of 7.7 points on multiple QA benchmarks.
Retrieval-Augmented Generation (RAG) enhances large language models (LLMs) by incorporating external knowledge, yet traditional single-round retrieval struggles with complex multi-step reasoning. Agentic RAG addresses this by enabling LLMs to dynamically decide when and what to retrieve, but current RL-based training methods suffer from sparse outcome rewards that discard intermediate signals and low sample efficiency where failed samples contribute nothing. We propose Search-P1, a framework that introduces path-centric reward shaping for agentic RAG training, comprising two key components: (1) Path-Centric Reward, which evaluates the structural quality of reasoning trajectories through order-agnostic step coverage and soft scoring that extracts learning signals even from failed samples, and (2) Dual-Track Path Scoring with offline-generated reference planners that assesses paths from both self-consistency and reference-alignment perspectives. Experiments on multiple QA benchmarks demonstrate that Search-P1 achieves significant improvements over Search-R1 and other strong baselines, with an average accuracy gain of 7.7 points.