• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2021, Vol. 57 ›› Issue (10): 42-49.doi: 10.3901/JME.2021.10.042

• 运载工程 • 上一篇    下一篇

扫码分享

基于环境识别的电动工程机械无人驾驶行走方法

林添良, 姚瑜, 许文杰, 付胜杰, 任好玲, 陈其怀   

  1. 华侨大学机电及自动化学院 厦门 361021
  • 收稿日期:2020-05-11 修回日期:2021-03-18 出版日期:2021-05-20 发布日期:2021-07-23
  • 通讯作者: 林添良(通信作者),男,1983年出生,博士,教授,博士研究生导师。主要研究方向为工程机械的电动化和智能化技术。E-mail:ltlkxl@163.com
  • 作者简介:姚瑜,女,1996年出生。主要研究方向为工程机械无人驾驶。E-mail:yaoyukxl@126.com
  • 基金资助:
    国家自然科学基金(51875218)、福建省高校产学研重大(2019H6015)、福建省STS计划配套(2018T3015)和福建省杰青(2018J06014)资助项目。

Driverless Walking Method of Electric Construction Machinery Based on Environment Recognition

LIN Tianliang, YAO Yu, XU Wenjie, FU Shengjie, REN Haoling, CHEN Qihuai   

  1. College of Mechanical Engineering and Automation, Huaqiao University, Xiamen 361021
  • Received:2020-05-11 Revised:2021-03-18 Online:2021-05-20 Published:2021-07-23

摘要: 无人驾驶技术是指车辆在任意环境和道路条件下,由无人驾驶系统控制车辆完成行驶任务的技术。工程机械作业环境恶劣,部分作业任务具有高度重复性,工程机械无人驾驶可有效降低驾驶员作业的风险,节约劳动力成本,提高作业效率。为实现电动工程机械的无人驾驶,针对电动工程机械的控制特性,采用基于摄像头的端到端决策系统。考虑到工程机械作业场景的特殊性,提出一种新型语义分割网络,该网络可实时融合由摄像头获取的空间特征,缩短图像处理时间,减少系统内存占用。通过在Udacity模拟器中测试发现,该系统可平稳完成模拟驾驶任务。搭建电动履带挖掘机无人驾驶综合试验平台,在非结构化道路上进行无人驾驶试验,试验结果表明,试验车可安全完成直线行驶、转弯任务,在小角度回正中具有较好的控制特性,为后续基于多摄像头的工程机械复杂工况识别与决策系统奠定基础。

关键词: 工程机械, 无人驾驶, 深度学习, 端到端, 语义分割

Abstract: Driverless technology is the technology that the vehicle is completely controlled by the driverless system to complete the driving task under any environment and road conditions. The operating environment of construction machinery is harsh, and some tasks are highly repetitive; driverless technology of construction machinery can effectively reduce the operation risk of drivers, save labor costs and improve operation efficiency. To realize the driverless electric construction machinery, according to the characteristics of construction machinery, an end-to-end decision system based on monocular camera is adopted. Considering the operation scenarios particularity, a semantic segmentation network is proposed that can fuse the spatial features acquired by the camera in real time, shorten the image processing time, and reduce the system memory occupation. Through testing in the Udacity, it is found that the system can smoothly complete the driving simulation task. To validate the system, an electric crawler excavator is chosen as the system’s experimental platform. The driverless experiments is carried out on an unstructured road. It can safely complete straight-line driving and turning tasks, and it has good control characteristics when returning to the center at a small angle, lays the foundation for the multi-camera construction machinery complex working condition recognition and decision system in addition.

Key words: construction machinery, driverless technology, deep learning, end-to-end, semantic segmentation

中图分类号: