跳到主要导航 跳到搜索 跳到主要内容

Deep Reinforcement Learning Algorithm and Simulation Verification Analysis for Automatic Control of Unmanned Vehicles

  • Yonghong Chen
  • , Yuxiang Zhang
  • , Jiaao Chen
  • , Junyu Zhao
  • , Ke Li*
  • , Lijing Wang
  • *此作品的通讯作者
  • Ltd
  • Beihang University

科研成果: 书/报告/会议事项章节会议稿件同行评审

摘要

This study conducted research mainly on the proven applicability of controlling the unmanned vehicle using a deep reinforcement learning algorithm and relative performance improvements. In specific, this study chose the AirSim platform developed by Microsoft as the simulation environment and conducted simulations mainly in the indoor parking lot Unreal 4 environment. In the simulations, the deep reinforcement learning method applied is Deep Q Networks for its effectiveness as well as simplicity. To improve the performance of the trained network, object detection methodology YOLO v3 is applied as the detection algorithm for the unmanned vehicle, and the network is improved using the output of object detection as its input to accelerate the training process. The implementation of the algorithms has efficiently proven the feasibility of using deep reinforcement learning agents for the unmanned vehicle in the project and the implementation of effective object detection.

源语言英语
主期刊名Man-Machine-Environment System Engineering - Proceedings of the 22nd International Conference on MMESE
编辑Shengzhao Long, Balbir S. Dhillon
出版商Springer Science and Business Media Deutschland GmbH
279-286
页数8
ISBN(印刷版)9789811947858
DOI
出版状态已出版 - 2023
活动22nd International Conference on Man-Machine-Environment System Engineering, MMESE 2022 - Beijing, 中国
期限: 21 10月 202223 10月 2022

出版系列

姓名Lecture Notes in Electrical Engineering
941 LNEE
ISSN(印刷版)1876-1100
ISSN(电子版)1876-1119

会议

会议22nd International Conference on Man-Machine-Environment System Engineering, MMESE 2022
国家/地区中国
Beijing
时期21/10/2223/10/22

指纹

探究 'Deep Reinforcement Learning Algorithm and Simulation Verification Analysis for Automatic Control of Unmanned Vehicles' 的科研主题。它们共同构成独一无二的指纹。

引用此