• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2026, Vol. 62 ›› Issue (17): 282-292.doi: 10.3901/JME.260716

• 特邀专栏:人形机器人与具身智能 • 上一篇    

扫码分享

基于改进YOLOv8的墙面砂浆缺陷修补智能双臂机器人关键技术研究

赵常屹, 白宇文, 赵润泽, 杨孔华, 刘春宝, 杨子赫   

  1. 吉林大学机械与航空航天工程学院 长春 130022
  • 收稿日期:2025-12-01 修回日期:2026-05-28 发布日期:2026-11-20
  • 作者简介:赵常屹,男,2000年出生,博士研究生。研究方向:具身智能机器人系统。E-mail:zhaocy23@mails.jlu.edu.cn;刘春宝(通信作者),男,1979年出生,博士,教授,博士研究生导师。主要研究方向为智能控制技术,仿生机器人。E-mail:liuchunbao@jlu.edu.cn
  • 基金资助:
    国家科技重大专项(2026ZD1608800)、国家自然科学基金(52405311)、博士后创新人才支持计划(BX20240139)和中国博士后科学基金(2024M761115)资助项目。

Key Technology Research on an Intelligent Dual-arm Robot for Wall Plaster Defect Repair Based on Improved YOLOv8

ZHAO Changyi, BAI Yuwen, ZHAO Runze, YANG Konghua, LIU Chunbao, YANG Zihe   

  1. School of Mechanical and Aerospace Engineering, Jilin University, Changchun 130022
  • Received:2025-12-01 Revised:2026-05-28 Published:2026-11-20

摘要: 传统墙面砂浆抹灰施工存在施工效率低、质量不稳定以及高度依赖人工经验进行缺陷检测和修补的问题,这不仅增加了劳动成本,还容易造成施工质量参差不齐。为解决这一问题,设计了一种基于改进YOLOv8的具身智能双臂抹灰机器人。所提出的改进YOLOv8模型在结构上引入了感受野注意力卷积(Receptive-field attention convolution, RFAConv)与多分支卷积模块(Diverse branch block, DBB),有效增强了模型对低对比度和弱纹理缺陷的感知能力,提高了缺陷检测的准确性。在墙面砂浆缺陷检测任务中,模型的mAP50达到了0.554,相较于基线算法提升了10%。机器人系统通过坐标系转换,实现了视觉检测结果的精准映射和缺陷的自主修补。实验结果表明,修补后的墙面平整度均值为10.53 mm,方差为0.63,验证了系统在复杂施工环境中的可行性和高效性。

关键词: 双臂操作机器人, YOLOv8, 墙面砂浆缺陷, 视觉检测, 缺陷修补, 自动化施工

Abstract: Traditional wall mortar plastering suffers from low construction efficiency, unstable quality, and a high dependence on manual experience for defect detection and repair. These issues not only increase labor costs but also often lead to inconsistent construction quality. To address this problem, this study designs a embodied intelligent dual-arm plastering robot based on an improved YOLOv8 model. The proposed improved YOLOv8 incorporates receptive-field attention convolution and a diverse branch block (DBB) module into its architecture, effectively enhancing the model’s ability to perceive low-contrast and weak-texture defects and improving defect detection accuracy. In the wall mortar defect detection task, the model achieved an mAP50 of 0.554, representing a 10% improvement over the baseline algorithm. The robotic system utilizes coordinate transformation to precisely map the visual detection results and autonomously repair defects. Experimental results show that the repaired wall achieved an average flatness of 10.53 mm with a variance of 0.63, demonstrating the feasibility and efficiency of the system in complex construction environments.

Key words: dual-arm manipulator robot, YOLOv8, wall mortar defect, visual detection, defect repair, automated construction

中图分类号: