Dynamic Decision Correction Framework Integrating Path Optimization and Incomplete Information Handling
Ziye Zhou, Yanbin Guo, Xin Tang, Kehan Li, Heng You · 2025
Despite significant advancements in multi-agent systems, solving the optimal path problem and addressing incomplete information remain challenging in multi-agent reinforcement learning (MARL). Traditional MARL algorithms often struggle with these two critical issues. To address these challenges, this paper proposes a novel framework called DDCF(Dynamic Decision Correction Framework) that integrates deep reinforcement learning (DRL) with Long Short-Term Memory (LSTM) networks to optimize the decision-making process for agents navigating complex environments. The first key innovation of this framework is the application of LSTM networks, which enable agents to learn optimal paths over time while considering dynamic, temporal dependencies within the environment. The second key innovation is the introduction of a dynamic decision-making approach, which effectively handles the problem of incomplete information by enabling agents to make decisions based on both observable and inferred states. This decision method ensures that agents can intelligently handle uncertain environments by incorporating both certain and fuzzy information. Experimental results demonstrate that the proposed framework significantly outperforms traditional methods in terms of path optimization and decision-making performance, providing a robust solution for agents in environments characterized by incomplete information and complex decision tasks.