Abstract
An overview of neuroevolutionary controller development is presented with an application to tractor-trailer backward path tracking. Controlling a tractortrailer vehicle driving in reverse is a difficult nonlinear control problem with widespread significance in the industry. Automated backward path tracking and docking has the ability to save substantial amounts of time and resources if implemented on a large scale. The presented work demonstrates both feedforward and recurrent neural network backward path tracking controllers for tractor-trailers evolved using a genetic algorithm. The example scenario demonstrates the utility of neuroevolved controllers for solving difficult nonlinear control problems. The neuroevolutionary techniques detailed in this work fall under the umbrella of reinforcement learning, and it is shown that the methods used for developing the tractor-trailer controller can be easily extended for solving more generic control problems.
