Aligning Progress and Feasibility: A Neuro-Symbolic Dual Memory Framework for Long-Horizon LLM Agents

Researchers propose a new Neuro-Symbolic Dual Memory Framework that addresses key limitations in large language models for long-horizon decision-making tasks. The framework separates semantic progress guidance from logical feasibility verification, significantly improving performance on complex AI tasks while reducing errors and inefficiencies.

• →LLMs struggle with long-horizon tasks due to Progress Drift and Feasibility Violation errors that existing single-paradigm approaches cannot effectively address.
• →The new framework uses dual memory systems: neural Progress Memory for semantic guidance and symbolic Feasibility Memory for logical validation.
• →Testing on ALFWorld, WebShop, and TextCraft showed significant performance improvements over existing competitive baselines.
• →The approach drastically reduces invalid action rates and average trajectory lengths in AI agent decision-making.
• →This represents a fundamental shift from single-paradigm to dual-paradigm approaches for complex AI agent tasks.