Model Predictive Control for Real-time Trajectory Generation in Racing

Abstract

Trajectory generation in autonomous racing is a problem that is a highly challenging task. A racing driver operates the vehicle at the absolute limit of its capabilities, in highly nonlinear areas.

Trajectory genration is often solved by dividing the problem into smaller parts; path planning and path following, where the path planning algorithm disregards vehicle dynamics. As such, the path planning algorithm must either make conservative estimates of the vehicle's capabilities, or risk planning an unattainable path.

This thesis aims to make use of optimal control in order to solve the two problems at once, allowing for the generation of trajectories at the limit of the vehicle's capabilities.

The controller has been implemented and tested in a real-time simulation environment, and has been proven to be both functional and performant, but also limited by its numerical conditioning. This simulation is shown at https://youtu.be/wVQK8j0KHKA.