Tackling complex decision-making? This research introduces GraphMDP, a novel decomposition tool designed to solve weakly-coupled Markov Decision Processes (MDPs). By using a predefined partition of the MDP, the tool constructs a directed graph to break down the global MDP into smaller, local MDPs that can be solved independently. Combining these local solutions yields an approximate solution for the larger global MDP. The focus is on efficiency and scalability for complex problems. The effectiveness of GraphMDP is demonstrated through its application in mobile robotics, where it achieves near-optimal solutions in considerably less time. Preliminary results also showcase a parallel implantation of the tool, indicating its potential for further performance improvements. This approach has significant implications for solving complex, real-world decision-making problems in various fields. Its capacity to reduce computational complexity while maintaining solution quality highlights its value in both theoretical and applied contexts.
Published in the International Journal on Artificial Intelligence Tools, this paper aligns directly with the journal's focus on artificial intelligence and its practical applications. The introduction of GraphMDP, a tool designed to solve complex Markov Decision Processes, demonstrates the journal's interest in innovative AI techniques and their use in solving real-world problems. The application of this tool to mobile robotics emphasizes its practical utility.