Comparação de controladores de rastreio de trajetória para veículos autônomos em um ambiente virtual sensoriado.

Abstract

In this project is presented the development of a simulation environment for autonomous

cars, in an architecture that permits the abstraction of sensors on the vehicle, the utiliza- tion, and creations of different scenarios, and motion controllers validation. The virtual

environment was developed on Simulink®, the vehicle dynamic simulation was done using the mathematical model of an Ackerman steering geometry vehicle and rear traction developed in this work. To visualize the scenarios and to the sensors get data similar to the one of real vehicles, was used a tool of Matlab to use the Unreal Engine to simulate the 3D scenarios, the scenarios were built by Driving Scenario Designer. Using the developed environment a comparison of three trajectory controllers to autonomous vehicles was performed: the Pure Pursuit, The Stanley’s method, and a Nonlinear Model Predictive Controle, to a Lemniscate approach trajectory. The controllers were analyzed based on integral squared error, total variation of the control signal, and the required calculation time. The results showed that the NLMPC is the best controller, regarding output error and control effort, but it needs a longer control signal calculation time. The Pure Pursuit controller was better than Stanley’s Method to the simulated situation of this work.