Model Predictive Control for Scalable Multi-Agent Systems: A Distributed Approach

Overview & Synopsis

This poster presents the initial stages of a research project focused on developing a Distributed Model Predictive Control (DMPC) framework for scalable and dynamically evolving multi-agent systems (MAS). The work addresses the limitations of traditional centralized control methods by enabling each agent to solve a local optimization problem subject to state, input, and environmental constraints. A key aspect of the framework is the decomposition of constraints into non-coupled (e.g., input limits and obstacle avoidance) and coupled constraints (e.g., inter-agent collision avoidance), allowing for parallelized computation and modular scalability. Initial results include the successful implementation of a local MPC controller on a single robotic platform for point stabilization in an obstacle-filled environment. Future work will incorporate interaction-aware constraints and extend the approach to multi-robot coordination. The poster was presented at the Journée Régionale des Doctorants en Automatique (JRDA) 2025 in Amiens, France.