3D视觉引导的机械臂力控曲面打磨方法

PDF下载HTML

Title:

Force-controlled robotic polishing of curved surfaces with 3D vision

作者:

万里红¹, 林杰¹, 刘娜², 张泽阳¹, 吴国栋¹, 蒋远东¹

1. 中原动力智能机器人有限公司, 河南 郑州 450046;
2. 上海理工大学 机器智能研究院, 上海 200093

Author(s):

WAN Lihong¹, LIN Jie¹, LIU Na², ZHANG Zeyang¹, WU Guodong¹, JIANG Yuandong¹

1. Origin Dynamics Intelligent Robot Co., Ltd, Zhengzhou 450046, China;
2. University of Shanghai for Science and Technology, Institute of Machine Intelligence, Shanghai 200093, China

关键词:

3D视觉; 六维力传感器; 力控; 曲面打磨; 分层控制; 曲率驱动; 导纳控制; 机械臂

Keywords:

3D vision; six-axis force sensor; force control; curved surface polish; hierarchical control; curvature-driven; admittance control; manipulator

分类号:

TP242

DOI:

10.11992/tis.202506024

摘要:

针对复杂曲面打磨中几何误差与接触力耦合导致的过磨、欠磨问题，本研究提出了一种视觉引导与六维力控协同的机械臂自适应打磨方法。系统通过3D结构光相机实时采集工件曲面点云数据，生成初始打磨轨迹；通过六维力传感器获取接触力/力矩信息，动态调整末端位姿以补偿曲面几何偏差。本研究采用分层控制架构，实现了视觉全局轨迹规划与力控局部微调协同；通过力矩反馈抑制工具侧滑，提升曲面贴合度；最后依据点云曲率在线调整虚拟刚度以避免过载。在针对复杂曲面工件打磨实验中，相较于仅视觉轨迹跟踪或者力控的方法，本研究显著降低了表面粗糙度Ra（roughness average）至0.8 μm，同时力跟踪误差减少了62%，并有效消除了由初始位姿偏差引起的脱离接触现象。

Abstract:

To address the issues of over-grinding and under-grinding caused by the interaction between geometric inaccuracies and contact forces in complex surface grinding, this study proposes a vision-guided and six-dimensional force-controlled collaborative adaptive grinding method for robotic arms. The system collects real-time point cloud data of the workpiece surface using a 3D structured light camera to generate the initial grinding trajectory, while a six-dimensional force sensor acquires contact force/moment information to dynamically adjust the end-effector pose for compensating surface geometric deviations. A hierarchical control architecture is adopted to achieve collaboration between global vision-based trajectory planning and local force-controlled fine-tuning. Torque feedback is utilized to suppress tool slippage and improve surface conformity. Additionally, virtual stiffness is adjusted online based on point cloud curvature to avoid overload. In grinding experiments on complex curved workpieces, compared to methods relying solely on visual trajectory tracking or force control, this study significantly reduces Ra to 0.8 μm, decreases force tracking error by 62%, and effectively eliminates contact loss caused by initial pose deviations.