Real-time decentralized model predictive control for cooperative multi-robot object transport: experimental validation

Abstract

This paper presents an experimental validation of a decentralized Model Predictive Control (MPC) framework for cooperative object transportation utilizing a multi-robot system consisting of two mobile robots. Each robot is a differential-drive robot that independently solves local constrained optimization problems while ensuring global coordination through joint-space coupling. The formulation explicitly captures nonlinear kinematics, revolute-prismatic joint dynamics, inter-robot constraints, and dynamic obstacle avoidance within a real-time optimization setting. Adaptive weighting of cost terms is employed to balance trajectory tracking and formation objectives under varying task demands. The framework is deployed on a physical testbed integrating vision-based pose estimation, sensor fusion via a Kalman filter, and a ROS 2 control infrastructure. Experiments across point-to-point, curvilinear, and obstacle-rich scenarios show accurate trajectory tracking, strict constraint satisfaction, and robustness to environmental uncertainties. These results substantiate decentralized constrained MPC with adaptive weights as a practical and scalable solution for real-time multi-robot cooperative transport along arbitrary reference paths.