Process Reward Models Meet Planning: Generating Precise and Scalable Datasets for Step-Level Rewards
For researchers building PRMs, this work provides a scalable, low-cost alternative to manual annotation that extends beyond math domains, though the approach is limited to problems expressible in PDDL.
The authors propose a scalable method to generate step-level reward datasets for Process Reward Models (PRMs) using planning problems expressed in PDDL, producing ~1M reasoning steps. Augmenting existing PRM training data with this PDDL-derived data improves performance on both mathematical and non-mathematical reasoning benchmarks.
Process Reward Models (PRMs) have emerged as a powerful tool for providing step-level feedback when evaluating the reasoning of Large Language Models (LLMs), which frequently produce chains of thought (CoTs) containing errors even when the final answer is correct. However, existing PRM datasets remain expensive to construct, prone to annotation errors, and predominantly limited to the mathematical domain. This work introduces a novel and scalable approach to PRM dataset generation based on planning logical problems expressed in the Planning Domain Definition Language (PDDL). Using this method, we generate a corpus of approximately one million reasoning steps across various PDDL domains and use it to train PRMs. Experimental results show that augmenting widely-used PRM training datasets with PDDL-derived data yields substantial improvements in both mathematical and non-mathematical reasoning, as demonstrated across multiple benchmarks. These findings indicate that planning problems constitute a scalable and effective resource for generating robust, precise, and fine-grained training data for PRMs, going beyond the classical mathematical sources that dominate this field.