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

2026-04-03

AI Summary

This work addresses the challenge of prolonged trial-and-error inefficiencies in long-horizon LLM agents operating in complex environments, where global progress drift and locally infeasible actions often lead to suboptimal behavior. The authors propose a neuro-symbolic dual-memory framework that explicitly decouples semantic progress guidance from logical feasibility verification. Specifically, a neural component learns task progression trajectories to inform high-level planning, while a symbolic component rigorously validates actions against formal constraints, synthesizing executable Python rules from failure experiences. Evaluated on ALFWorld, WebShop, and TextCraft, the approach significantly outperforms existing baselines, markedly reducing both the rate of invalid actions and average trajectory length.

Abstract

Large language models (LLMs) have demonstrated strong potential in long-horizon decision-making tasks, such as embodied manipulation and web interaction. However, agents frequently struggle with endless trial-and-error loops or deviate from the main objective in complex environments. We attribute these failures to two fundamental errors: global Progress Drift and local Feasibility Violation. Existing methods typically attempt to address both issues simultaneously using a single paradigm. However, these two challenges are fundamentally distinct: the former relies on fuzzy semantic planning, while the latter demands strict logical constraints and state validation. The inherent limitations of such a single-paradigm approach pose a fundamental challenge for existing models in handling long-horizon tasks. Motivated by this insight, we propose a Neuro-Symbolic Dual Memory Framework that explicitly decouples semantic progress guidance from logical feasibility verification. Specifically, during the inference phase, the framework invokes both memory mechanisms synchronously: on one hand, a neural-network-based Progress Memory extracts semantic blueprints from successful trajectories to guide global task advancement; on the other hand, a symbolic-logic-based Feasibility Memory utilizes executable Python verification functions synthesized from failed transitions to perform strict logical validation. Experiments demonstrate that this method significantly outperforms existing competitive baselines on ALFWorld, WebShop, and TextCraft, while drastically reducing the invalid action rate and average trajectory length.

Problem

• Progress Drift
• Feasibility Violation
• Long-Horizon Decision-Making
• LLM Agents
• Neuro-Symbolic

Innovation

• Neuro-Symbolic
• Dual Memory
• Progress Drift
• Feasibility Violation
• Long-Horizon Planning