Local Navigation and Obstacle Avoidance for an Agricultural Tractor With Nonlinear Model Predictive Control

An important step in automating agricultural field work is to automate the navigation and driving of the agricultural vehicles. We present a nonlinear model predictive control (NMPC)-based path tracking and obstacle avoidance approach for agricultural front-wheel-steered tractors. The prediction model in the NMPC is based on the kinematic model of an Ackermann vehicle and the actuators for speed and steering are modeled as first-order systems without time delay. Both the path tracking and the obstacle avoidance are based on the cost function formulation that penalizes the cross track error (xte) and closeness to the obstacles. Both simulation and real-life experiments were carried out to evaluate the capabilities of the controller. Multiple test cases were used to assess different aspects of the controller behavior. We show that the NMPC achieves a xte of less than 0.05m on straight sections of the reference path, and is able to avoid simple obstacles, also when multiple obstacles need to be considered at the same time. We also show that the computational cost is feasible for real-time applications on the field, even with a long prediction horizon in the NMPC.