SRMP-S: Self-Refining Multi-Planner Strategy for Long-Horizon Robotic Tasks

Abstract

Enabling robotic agents to perform complex long-horizon tasks has been a long-standing goal in robotics and artificial intelligence (AI). Existing Large Language Models (LLMs)-based symbolic planners auto-generate Planning Domain Definition Language (PDDL) specifications to bridge language and formal reasoning, but they rely on a single open-loop planner and lack autonomous error diagnosis or model refinement, limiting robustness in long-horizon tasks. To address these issues, we propose Self-Refining Multi-Planner Strategy (SRMP-S), a closed-loop framework that integrates the semantic flexibility of LLMs with the rigor of symbolic planning through two key innovations. Firstly, SRMP-S employs a multi-strategy planner ensemble, concurrently executing multiple heterogeneous symbolic planners to broaden first-attempt solvability and produce high-fidelity, cross-validated failure diagnostics. Secondly, it introduces a self-refinement mechanism that automatically converts these diagnostics into structured natural language (NL) feedback to iteratively refine the LLMs-generated PDDL models, enabling autonomous recovery from initial modeling inaccuracies. Experimental results demonstrate that SRMP-S achieves state-of-the-art performance across multiple domains, outperforming existing LLM-based planners.