1 of 1

Results:

• Reduce model mismatch by as much as 75% using Gaussian Processes Regression.
• Improve lap-time by 42% compared to previous control architecture.
• Improve lap-time by 12% during the course of a 10-lap event.

Objectives:

​

• Improve the Motion Planning and Control Pipeline of IST’s Autonomous Formula Student Racing Prototype
• Exploit Existing Data - Pre-recorded and Collected Online to Enhance a Model-Based Controller Performance

Learning to Drive Autonomously a Race Car

​

João Gomes de Oliveira Pinho

Dissertação para a obtenção do Grau de Mestre em Engenharia Mecânica (Setembro 2021)

Orientadores: Prof. Pedro Urbano Lima e Prof. Miguel Ayala Botto

Emails: joao.g.pinho@tecnico.ulisboa.pt; pedro.lima@tecnico.ulisboa.pt; ayalabotto@tecnico.ulisboa.pt

Learning of the Terminal Constraints

Learning of the Model Error

Offline

Online

75%