Design and experimental validation of a robust model predictive control for the optimal trajectory tracking of a small-scale autonomous bulldozer

Abstract

Trajectory tracking of an unmanned ground vehicle (UGV) is essential due to its extensive construction, agriculture, and military applications. In this paper, we propose an efficient, robust model predictive control (RMPC) for the trajectory tracking of a small-scale autonomous bulldozer in the presence of perturbations by unknown but bounded disturbances. The proposed RMPC is designed by considering a linearised tracking error-based model combined with a feed-forward and optimal control action to achieve the proposed trajectory. The presence of a corrective feedback controller as a time-varying finite-time linear quadratic regulator (LQR) suppresses the uncertainties acting on the real system by regulating around the nominal system. Pose estimation, required for control feedback, is based on sensor data fusion performed by an extended Kalman filter (EKF) map-based localiser, which processes inertial measurement unit (IMU) and light detection and ranging (LiDAR) measurements. Experiments are performed using a real robot (Husky A200) to validate the proposed control scheme’s performance. The experimental results show that the proposed controller can safely track target trajectories with low processing time, small tracking errors, and smooth control actions. Finally, the proposed control scheme is compared with related techniques and outperforms them in tracking accuracy.