Nonlinear model predictive control for autonomous driving in terrain

Forestry machines used nowadays function in a wide variety of terrains. In order to make these vehicles operate autonomously advanced control methods are needed. However, current research related to autonomous driving generally assumes flat terrain in the control as well as in the estimation. Therefore concerning this matter, this paper addresses potential solutions by investigating and applying nonlinear model predictive control (NMPC) for autonomous driving in uneven terrain. This paper proposes a hybrid model derived from the dynamic six-degrees-of-freedom (6-DOF) model for motion control purposes. The developed NMPC method utilises a path tracking approach and aims to minimise the time-independent tracking error between the position of the vehicle and path by utilising the proposed hybrid model and three-dimensional (3D) terrain map. The effectiveness of the predictive controller is tested using three different test paths and terrains. An all-terrain electric vehicle (ATV) called Polaris is utilised to test and confirm the functionality of the control method. In addition, the paper proposes a rollover avoidance method and tests it in simulation environment. The method aims to lower the vehicle speed in the presence of high roll angles. The results from the actual tests with the implementation of the NMPC method indicate that accurate path-tracking results can be obtained with the proposed controller in the test paths used in this study with the tracking errors being 0.11 m, 0.07 m and 0.1 m.