作者：李世杰1,2, 左治江1,2, 李涵1,2
作者单位：1. 江汉大学 精细爆破国家重点实验室, 湖北 武汉 430056;
2. 爆破工程湖北省重点实验室, 湖北 武汉 430056
关键词：机器人抓取系统;计算机视觉;深度学习;目标检测
摘要：
针对传统人工示教机器人抓取系统智能化程度低、自适应性差、系统可移植性差等问题，提出一种基于深度学习的机器人抓取系统。搭建由UR3机械臂，Gigabyte Z370N WIFI嵌入式开发板，深度相机等组成的硬件系统，通过基于深度学习的目标检测获取目标物像素坐标，进行视觉系统的定位研究将像素坐标映射到世界坐标系并在ROS系统中进行机械臂建模与控制，完成对目标物的抓取。针对由于嵌入式硬件平台算力欠缺导致目标检测算法难以部署的问题，从目标检测精度、实时性两方面将工业界应用广泛的YOLO系列算法进行对比研究，探究基于深度学习的目标检测算法在算力欠缺的嵌入式机器人抓取系统中的应用。实验结果表明，基于计算机视觉的机器人抓取系统目标识别精度高，智能化程度和鲁棒性高，具有良好的可移植性，有一定的推广应用价值。
Design of robot grasping system based on deep learning
LI Shijie1,2, ZUO Zhijiang1,2, LI Han1,2
1. State Key Laboratory of Percision Blasting, Jianghan University, Wuhan 430056, China;
2. Hubei Key Laboratory of Blasting Engineering, Wuhan 430056, China
Abstract: Aiming at the problems of low intelligence, poor adaptability and poor portability of traditional manual teaching robot grasping system, a robot grasping system based on deep learning is proposed. A hardware system consisting of UR3 manipulator, Gigabyte Z370N WIFI embedded development board and depth camera is built. The pixel coordinates of the target are obtained through the object detection based on deep learning. The pixel coordinates are mapped to the world coordinate system through the positioning research of the visual system, and the manipulator modeling and control are carried out in the ROS system to complete the capture of the target. Aiming at the problem that the target detection algorithm is difficult to deploy due to the lack of computing power in the embedded hardware platform, the YOLO series algorithms widely used in the industry are compared and studied from the aspects of target detection accuracy and real-time performance, and the application of the target detection algorithm based on deep learning in the embedded robot grasping system with insufficient computing power is explored. The experimental results show that the robot grasping system based on computer vision has high target recognition accuracy, high intelligence and robustness, good portability, and great application value.
Keywords: robot grasping system;computer vision;deep learning;object detection
2023, 49(5):129-136 收稿日期: 2022-06-13;收到修改稿日期: 2022-08-06
基金项目: 国家重点研发计划资助项目(2021YFB2301004)；爆破工程湖北省重点实验室2021年度开放基金(BL2021-16)
作者简介: 李世杰(1997-),男,江苏连云港市人,硕士研究生,专业方向:计算机视觉,机器人抓取
参考文献
[1] 姜伟平, 孟子晗, 嵇志康, 等. 基于视觉伺服的机器人辅助医用显示器可视角检测系统设计[J]. 中国测试, 2022, 48(5): 128-133
[2] AUDE B, DANICA K. Trends and challenges in robot manipulation[J]. Science, 2019, 364(6446): 1149
[3] 秦方博, 徐德. 机器人操作技能模型综述[J]. 自动化学报, 2019, 45(8): 1401-1418
[4] 刘乃军, 鲁涛, 蔡莹皓, 等. 机器人操作技能学习方法综述[J]. 自动化学报, 2019, 45(3): 458-470
[5] 谢宇珅, 吴青聪, 陈柏, 等. 基于单目视觉的移动机械臂抓取作业方法研究[J]. 机电工程, 2019, 36(1): 71-76
[6] 郭圆辉, 王旭彬, 韩毅. 基于RGB-D相机的机械臂抓取系统设计[J]. 河南科技, 2021, 40(29): 11-14
[7] 杜玉俊, 吴盘龙. 基于ROS的七自由度机械臂抓取系统研究[J]. 测控技术, 2021, 40(6): 39-45
[8] 张青春, 何孝慈, 姚胜, 等. 基于ROS机器人的相机与激光雷达融合技术研究[J]. 中国测试, 2021, 47(12): 120-123
[9] CAI Z X, LIANG Z, REN J. MRDRL-ROS: a multi robot deep reinforcement learning platform based on robot operating system[J]. Journal of Physics: Conference Series, 2021, 2113(1): 012086
[10] ERDOMU A K, KARACA D Z T M, YAYAN U. Robot operating system compatible mobile robots for education and research[C]//International Conference on Robotics and Automation, 2021.
[11] REDMON J, FARHADI A. YOLOv3: an incremental improvement[R]. arXiv e-prints, 2018.
[12] 段仲静, 李少波, 胡建军, 等. 深度学习目标检测方法及其主流框架综述[J]. 激光与光电子学进展, 2020, 57(12): 59-74
[13] ZHANG S, CHAI L, JIN L. Vehicle detection in UAV aerial images based on improved YOLOv3[C]//2020 IEEE International Conference on Networking, Sensing and Control (ICNSC). IEEE, 2020.
[14] 邓佳桐, 程志江, 叶浩劼. 改进YOLOv3的多模态融合行人检测算法[J]. 中国测试, 2022, 48(5): 108-115
[15] 刘月祺, 辛大欣, 华瑾. 基于ROS的六自由度机械臂运动规划[J]. 自动化与仪器仪表, 2020(2): 20-22.
[16] 李柯泉, 陈燕, 刘佳晨, 等. 基于深度学习的目标检测算法综述[J]. 计算机工程, 2022, 48(7): 1-12
