Towards Reliable and Robust LLM Planning: Symbolic Feedback-Driven Iterative Self-Refinement Framework

Short summary

This paper proposes a symbolic feedback-driven framework to improve LLM reliability in long-horizon planning. The approach uses natural language prompting to map logical symbols, a symbolic verifier for error correction, and a plan recognizer for goal feasibility. Empirical results show consistent improvements in planning task feasibility and correctness.

• •Introduces symbolic feedback-driven framework for LLM planning robustness
• •Maps logical symbols to natural language and corrects errors via symbolic verifier
• •Demonstrates improved feasibility and correctness in long-horizon planning tasks

Generated with AI, which can make mistakes.