Application of deep learning in the maneuvering control of a formula SAE vehicle

• Grant number: 18/11528-0
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: September 01, 2018
• End date: July 31, 2019
• Field of knowledge: Engineering - Electrical Engineering
• Principal Investigator: Leticia Rittner
• Grantee: Daniel Duck
• Host Institution: Faculdade de Engenharia Elétrica e de Computação (FEEC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil

Abstract

Recent advances in computational capacity led to the application of Artificial Intelligence in several activities, such as health, mobility, natural language recognition, law, computer vision and a wide range of research fields. One of its most famous application is the use of artificial intelligence to develop autonomous carsIts large-scale application has the potential to solve several of the modern and future urban challenges, such as pollution and traffic, which both hinder the urban growth and compromise the residents' quality of life. Two major obstacles that still hinder the development of fully autonomous urban vehicles are: the unpredictability of other drivers, pedestrians, cyclists and other traffic agents' behavior; and the possibility of fatal consequences of an error committed by the system. In order to stimulate the development of autonomous vehicles without risking people's lives, or having to deal with the vehicle's interaction with unpredictable traffic agents, SAE (Society of Automotive Engineers) has created the Formula Student Driverless (FSD) competition, in which university students must develop a racecar capable of completing a circuit with cones at both sides. Based on the success reported by similar projects, this research project aims to study the application of neural networks to learn the maneuvering control of a vehicle in a track delimited by cones on both sides, similarly to the FSD competition, based on a frontal camera. In order to do this, we will acquire and process the data of a car driving in a closed circuit and then train several neural networks to compare their performances in reproducing the driver's maneuvering.