连续两段式果蔬采摘机械臂视觉伺服控制方法研究

Research on Continuous Two-stage Visual Servo Control for Robotic Arm in Fruit and Vegetable Harvesting

摘要:

为了提高机械臂采摘果蔬的速度和稳定性，本文结合单目相机与激光测距传感器提出一种连续两段式的基于图像的视觉伺服控制方法。检测环节采用 YOLO (You only look once) v5s 网络实现目标果实的检测，进而对检测目标做阈值化分割获得目标区域。视觉伺服过程中以目标果实的中心点为图像特征，结合 PD (Proportional derivative) 控制建立机械臂关节的速度控制率。视觉伺服过程为对准阶段和接近阶段，两阶段的目标深度信息分别源自目标在图像中的半径估计和激光测距传感器的测量。为保证两阶段切换时纵向速度的连续，使用动态权重方法在过渡区域进行速度平滑。采用基于动态阻尼最小二乘法进行关节速度求解，避免了机械臂在接近奇异位形区域时的失控。开展机械臂采摘苹果试验，试验结果表明，提出的连续两段式控制方法，使机械臂的运动时间从 4.52s 降低至 2.56s, 无扰动最大绝对定位误差 4.00mm, 且能够稳定经过机械臂奇异位形区域，验证了该方法较高的速度和较强的鲁棒性。

Abstract:

In order to improve the picking speed and stability of picking manipulator, a continuous two-stage image-based visual servo control method combined monocular camera and laser distance sensor was proposed. The YOLOv5s network was used to detect the target fruit, and then threshold segmentation was performed on the detected target to obtain the target area. In the visual servo process, the center point of the target fruit was taken as the image feature, and the velocity control law of the manipulator joint was established combined with PD control. The visual servo process was divided into alignment stage and approach stage, and the target depth information in the two stages was obtained from the radius estimation of the target in the image and the measurement data of the laser ranging sensor, respectively. In order to ensure the continuity of longitudinal velocity during phase switching, the dynamic weight method was used to smooth the velocity in the transition zone. The joint velocity was calculated by the dynamic damping least squares method to avoid the out-of-control of the manipulator when approaching the singular configuration area. The apple picking experiment was carried out. The results showed that the proposed continuous two-stage control method reduced the movement time of the manipulator from 4.52s to 2.56s, the maximum absolute positioning error without disturbance was 4.00mm, and it could stably pass through the singular configuration area of the manipulator, which verified the high speed and strong robustness of the method.