Data-driven Model and Trajectory Tracking SMC for a UGV system

One of the most relevant features for the design of control algorithms is to show their robustness. For this reason, the main objective of this thesis has been the design of a robust controller for the target Unmanned Ground Vehicle (UGV) by means of Sliding Mode Control (SMC) technique, a well-known control strategy that provides this desired feature. To reach such objective, a new firmware for the UGV has been designed and a data-driven model has been built. The classical controller design procedure requires an initial characterization of a model which should be quite realistic but light, in order to make fast and reliable simulations. In this case, the provided model is a data-driven one, making possible the use of a kinematic model only of the UGV, keeping out dynamical modeling. A System Identification (SI) procedure has been necessary and has involved the collection of data, pre-processing, and finally model construction and validation; this phase confirmed the effectiveness of SI techniques and models, and was essential to refine the nonlinear auto regressive exogenous (NARX) model used for controller design and simulations purposes. In the second phase of this thesis, a new firmware for the UGV has been implemented, introducing system states through a finite state machine (FSM); its structure is inspired from some well-known frameworks related to UAVs. The firmware provides also Ethernet communications with another board, where some ROS nodes run. Those ROS nodes are the designed controller (Trajectory Tracking SMC), the reference generator (Artificial Potential Fields technique, APF), and the navigation one (Extended Kalman Filter, EKF). An important result of such firmware structure is the modularity: it makes possible to test different control algorithms just by substituting the ROS node, but also the APF and EKF node could be substituted to test different navigation and reference generator algorithms. This modularity feature makes also easier the replacement of sensing and/or actuation components. Finally, the new desired controller has been designed and tested in MATLAB/Simulink environment, where good performances have been observed in simulations, confirming the reaching of a robust control law (compared to the previous Proportional-Integral-Derivative controllers). Some tuning of the controller gains has been made and the block diagram of the controller subsystem has been translated into Python code manually, implementing a new ROS node. At last, the goal was to test the physical effectiveness of the controller through laboratory experiments in different scenarios; results proved good performance in the proposed scenarios, where the time of target goal reaching has been reduced and the overall behavior of the UGV during experiments has been considered satisfying.