Nonlinear model predictive control optimization for autonomous mobile robots

In the Agent-Target Coordination field, one of the most researched area is the coordination of a group of heterogeneous mobile agents in order to accomplish advanced tasks. Thanks to the accessibility and the improvement of Unmanned Aerial Vehicle (UAV) and Unmanned Ground Vehicle (UGV) over the last decades, their use in exploration, research and industrial cooperative applications is increasing. However, solving challenging tasks, such as {search\&rescue} and environmental monitoring, demands the application of control laws that require high performance computational systems. Despite the components miniaturization, the complexity of developing light-weight but performing processors has lead the growth of cloud-computing. In this thesis, it is addressed the problem of driving a UGV to follow an UAV in order to set up a landing scenario, while dealing with computational resources allocation. Specifically, the target trajectory is not know in advance by the agent and the only source of information concerning the poses and velocities of both vehicles comes from the camera attached to the external computational node. To solve this problem, it is proposed a cascade of control techniques based on the Model Predictive Control (MPC) and Gaussian Process Regression (GPR) approaches. The Model Predictive Control controller is devoted to solving the Agent-Target Coordination problem by driving the UGV under the aerial vehicle. The GPR module, instead, is dedicated to predicting the computational effort of the controller, to providing the MPC control invariant $N$ and to allocating the computation of the Model Predictive Control solution locally or on the external node. Simulation results on Matlab are presented in order to illustrate and validate the proposed approach.