Decision-making of autonomous driving based on reinforcement learning

• Main Author: Lee, Julia Hui Hui
• Other Authors: Lyu Chen
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Mechanical engineering::Motor vehicles
• Online Access: https://hdl.handle.net/10356/167670
• Institution: Nanyang Technological University

Autonomous driving (AD) technology has garnered significant interest in recent years due to its potential to transform transportation. However, despite advancements in AD technologies, current vehicles on the road are only partially autonomous, with limited autonomous features. Among the different stages in the AD pipeline, the decision-making process, particularly the prediction stage, has received relatively less attention and development compared to other modules. This is concerning as the decision-making stage is crucial for the safe and efficient operation of autonomous vehicles in any environment. Although there are existing studies on End-to-End Autonomous Driving, it does not provide enough insights into the selection and evaluation of reinforcement learning (RL) models for decision-making in autonomous driving tasks. Therefore, this paper is intended to investigate and compare the performance of two commonly used RL models, Proximal Policy Optimization (PPO) and Deep Q-Network (DQN), in a simulated autonomous driving scenario. The models are evaluated based on quantitative performance metrics such as collision rate, goal reached rate, and average distance covered, as well as qualitative behaviors observed during simulation runs.