Image-based Reinforcement Learning for Autonomous Vehicles

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering

Final Year Thesis Oral Defense

Title: "Image-based Reinforcement Learning for Autonomous Vehicles"

by

TANG Yiu Ting

Abstract:

An autonomous vehicle guidance system is developed using a recently 
proposed actor-critic and model-free algorithm[9]. A range of autonomous 
agents powered by different convolutional neural networks (CNNs) has been 
investigated and their performance on keeping themselves in lane in the 
TORCS[10] simulator is also compared. It is found that several types of 
CNNs are able to achieve good performance on the road and some of them 
outperform the others. The results provide a direction of choosing the 
architecture for the vision-based controllers of self-driving cars.

Date                 : 4 May 2017 (Thu)

Time                 : 15:40 - 16:20

Venue                : 2127C (via lift 19)

Advisor              : Prof. Chi-Keung TANG

2nd Reader           : Dr. Desmond TSOI