Lane-changing Decision-making using Single-step Deep Q Network

Yizhou Song, Kaisheng Huang, Wei Zhong

Abstract

The lane-changing decision-making is a great challenge in autonomous driving system, especially to judge the feasibility of lane-changing due to the randomness and complexity of surrounding traffic participants. Reinforcement learning has been shown to outperform many rule-based algorithm for some complex systems. In this paper, the single-step deep Q network algorithm is proposed by combining single-step reinforcement learning and deep Q network, and it is applied to judge the feasibility of lane-changing for autonomous vehicle. In a real-world-like and random traffic environment built in Carmaker, the trained agent can make correct judgment about the feasibility of lane-changing. Comparing the single-step deep Q network with the general deep Q network, although the general deep Q network can converge, there are still collisions, and the agent trained by single-step deep Q network is absolutely safe.

Download


Paper Citation


in Harvard Style

Song Y., Huang K. and Zhong W. (2020). Lane-changing Decision-making using Single-step Deep Q Network.In Proceedings of the International Symposium on Frontiers of Intelligent Transport System - Volume 1: FITS, ISBN 978-989-758-465-7, pages 25-32. DOI: 10.5220/0010009600250032


in Bibtex Style

@conference{fits20,
author={Yizhou Song and Kaisheng Huang and Wei Zhong},
title={Lane-changing Decision-making using Single-step Deep Q Network},
booktitle={Proceedings of the International Symposium on Frontiers of Intelligent Transport System - Volume 1: FITS,},
year={2020},
pages={25-32},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0010009600250032},
isbn={978-989-758-465-7},
}


in EndNote Style

TY - CONF

JO - Proceedings of the International Symposium on Frontiers of Intelligent Transport System - Volume 1: FITS,
TI - Lane-changing Decision-making using Single-step Deep Q Network
SN - 978-989-758-465-7
AU - Song Y.
AU - Huang K.
AU - Zhong W.
PY - 2020
SP - 25
EP - 32
DO - 10.5220/0010009600250032