Distributed Model Predictive Control for Multi-Agent Systems: Conception and Design for Dynamic Routing Task

This paper presents a sequential distributed model predictive control (DMPC) solution for multi-robot navigation. The method targets scenarios where multiple agents must reach designated goals while avoiding collisions in a shared environ- ment. Unlike centralized MPC, which suffers from scalability and communication bottlenecks, the proposed solution decomposes the problem into local optimizations solved sequentially in a fixed cyclic order. Collision avoidance is enforced through dy- namically activated constraints, which are introduced only when predicted inter-agent distances fall below a threshold, thereby reducing conservatism while ensuring safety. The solutionis implemented in ROS 2 using the ChoiRbot library and extended to support unicycle dynamics, two-dimensional motion, and soft Manhattan distance constraints. Simulation results in Gazebo with TurtleBot3 robots demonstrate collision-free convergence in both sparse two-agent and dense four-agent intersection scenarios. These findings highlight the potential of sequential DMPC as a scalable and communication-efficient solution for safe multi-robot coordination.