A Nonlinear Model Predictive Control Algorithm for an Unmanned Ground Vehicle on Variable Terrain

Abstract

This thesis presents a Nonlinear Model Predictive Controller (NMPC) for an Unmanned Ground Vehicle (UGV) capable of controlling the vehicle over both smooth and rough terrain using measurements from GPS and a Light Detection And Ranging (LiDAR) unit equipped to the vehicle. Linear Model Predictive Control (MPC) and NMPC have become more widely used to control dynamic systems as computers have become more capable of handling the computational expense required by model predictive control. Though the use of NMPC rather than linear MPC creates an additional computational expense, NMPC allows for path planning in addition to control of the vehicle. This is particularly advantageous in scenarios in which the UGV is traversing terrain that contains obstacles of which the vehicle has no a priori knowledge.

Rough, off-road terrain contains multiple hazards for an UGV. In this thesis, 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. The NMPC presented in this thesis is designed to mitigate this risk of rollover. Simulations of the NMPC in several different scenarios are presented, as well as results from experimental implementation of the NMPC on a test vehicle. Results from simulation and experimental implementation are provided that show the NMPC is able to navigate a UGV around obstacles to a target location without requiring the use of a priori knowledge of terrain and obstacles.