Aligning Progress and Feasibility: A Neuro-Symbolic Dual Memory Framework for Long-Horizon LLM Agents
Summary: arXiv:2604.02734v1 Announce Type: new
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.
Introduction
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.
Proposed Solution
Motivated by this insight, we propose a Neuro-Symbolic Dual Memory Framework that explicitly decouples semantic progress guidance from logical feasibility verification. This novel approach allows for a more effective handling of long-horizon tasks by utilizing two distinct memory mechanisms during the inference phase:
- Progress Memory: A neural-network-based mechanism that extracts semantic blueprints from successful trajectories to guide global task advancement.
- Feasibility Memory: A symbolic-logic-based mechanism that utilizes executable Python verification functions synthesized from failed transitions to perform strict logical validation.
Experimental Results
Our experiments demonstrate that this method significantly outperforms existing competitive baselines on various benchmarks, including ALFWorld, WebShop, and TextCraft. The results reveal substantial improvements in the following areas:
- Reduction in the invalid action rate, enhancing the reliability of the agents.
- Decrease in average trajectory length, leading to more efficient decision-making processes.
- Improved overall performance across complex decision-making tasks.
Conclusion
The Neuro-Symbolic Dual Memory Framework presents a significant advancement in the field of long-horizon decision-making. By effectively decoupling the processes of semantic guidance and logical validation, our approach addresses the fundamental issues of Progress Drift and Feasibility Violation. This innovative framework not only enhances the capabilities of large language models but also provides a robust solution for complex environments where decision-making is critical.
Future Work
Looking forward, we aim to further refine this framework by incorporating additional layers of reasoning and exploring its applicability across diverse domains. The potential for real-world applications remains vast, and continued research in this area could pave the way for more intelligent and capable LLM agents in the future.