Obstacle Avoidance of an Unmanned Ground Vehicle using a Combined Approach of Model Predictive Control and Proportional Navigation

Abstract

This thesis presents a new approach for the guidance and control of a UGV (Unmanned Ground Vehicle). A special focus was placed on moving obstacles that interfere with the planned path of the vehicle, this is due to the fact that the majority of obstacle avoidance research has been completed on stationary objects. An obstacle avoidance algorithm was developed using an integrated system involving Proportional Navigation (Pro-Nav) and a Nonlinear Model Predictive Controller (NMPC). An obstacle avoidance variant of the ideal proportional navigation law generates command lateral accelerations to avoid obstacles, while the NMPC is used to track the reference trajectory given by the Pro-Nav. The NMPC utilizes a lateral vehicle dynamic model along with a nonlinear tire model in order to issue control inputs. In this application an obstacle avoidance algorithm can take over the control of a vehicle until the obstacle is no longer a threat. Another application of a Pro-Nav and NMPC algorithm was tested for leader/follower situations. The performance of the leader/follower and obstacle avoidance algorithm is evaluated through different simulations.

Simulation of the performance of the PNCAG and NMPC algorithm was conducted us ing two different simulation environments; MATLAB and Simulink Vehicle Dynamics Block set. The MATLAB simulation validated the algorithm showing that it could be used to accomplish obstacle avoidance. With the algorithm shown to be effective, it was placed into the Vehicle Dynamics Blockset. The Vehicle Dynamics Blockset provided a higher fidelity vehicle model to provide a more realistic simulation environment. In addition to obstacle avoidance, simulation results verified the performance of a modified version of the PNCAG and NMPC algorithm in a leader/follower scenario. The results show, the algorithm handled the leader/follower and collision avoidance with reasonable error. Overall the algorithm was also able to follow a lead vehicle throughout a double lane change as well as avoid collision with a moving obstacle in four different scenarios.