Nonlinear Model Predictive Controller for an Unmanned Ground Vehicle on Variable Terrain

Rough, off-road terrain contains multiple hazards for an unmanned ground vehicle (UGV). In this paper, hazards are classified into three groups: obstacles, rough traversable terrain, and rough untraversable terrain. These three types of hazards create a rollover risk for a UGV. A nonlinear model predictive controller (NMPC) that is capable of navigating a UGV through these hazards is presented. The control algorithm features a nonlinear tire model which more accurately captures the dynamics of the UGV when compared to a linearized tire model, and has a fast enough run time for real time implementation. On an actual vehicle, the UGV is assumed to be equipped with a perception based sensor, such as a Light Detection And Ranging (LiDAR) unit, to provide information of the terrain roughness, grade, and elevation. This information is used by the NMPC to safely control the vehicle to a target location. However, for the purposes of this paper, control inputs and terrain are simulated in Car-Sim [1], and the feasibility of real time implementation is investigated.