Mechanical multi-agent maneuvering using noncooperative DMPC

Abstract

Distributed Model Predictive Control is used to coordinate agents in multi-agent systems by managing set-points and coupling constraints. The coordination of multi-agent systems concept regards all type of control algorithms dependent on information interchange between subsystems. The control algorithms are developed to solve a series of static optimization problems with nonlinear coupling constraints by means of a typical receding horizon policy applied in predictive control design. For distributed scenarios, the motion of each agent is determined by the distributed algorithm as function of the information shared with the other agents making the individual behavior implicitly dependent on a global outcome or cost. The control algorithm is used to maneuver dynamically decoupled mechanical agents in a two dimensional scenario with obstacles avoidance. The found solution is meaningful to realize how Predictive Control can be an alternative to other solutions obtained through Dynamic Games, where the agents play an important role, in a strategic space, as game players or Computational Intelligence technique, where the agents present a self-organized behavior. Hence, the developed algorithm is useful to maneuver unmanned vehicles in mazes, formations and also for collision avoidance.