Implementation of Robust Decentralized Model Predictive Control

Abstract

An algorithm for robust Decentralized Model Predictive Control (DMPC) is extended to explicitly account for delays in the system, arising from both the computation of each control optimization and the communication between subsystems. The algorithm is demonstrated in hardware, using wheeled robot vehicles to emulate Unmanned Aerial Vehicles. The behaviors of the vehicles are coupled by the need to avoid collisions, exercising the ability of the controller to make cooperative decisions. The DMPC algorithm divides the centralized team optimization problem into smaller sub-problems, each solving for the path of only one vehicle. The vehicles exchange plans by communication, but do not iterate on their intentions, and robust satisfaction of the coupled constraints is guaranteed under the assumption of a bounded disturbance. Experimental results are presented to verify that the modified algorithm operates in the presence of realistic computation and communication delays.