Real-time Visual Perception for Rapid Grasping of Underwater Delta Robot

Underwater autonomous grasping of seafood is a very meaningful field in robot research. However, the grasping efficiency of existing underwater grasping robots is very low. In this paper, a Delta robot for underwater capture is built, and a visual perception framework suitable for the rapid grasping of the manipulator is proposed. Firstly, we use YOLOv5 to initially locate the target, and then use the stereo camera to obtain the depth and calculate the spatial coordinates. Secondly, we established a refraction model for the waterproof housing, based on the problem that the target may be occluded, it is proposed to install a camera on the moving platform of the Delta manipulator to solve the occlusion problem. The vision system transmits the spatial coordinates of the target to the inverse kinematics controller. Finally, the Delta manipulator is built in the air and this framework is applied to the manipulator to implement rapid grasping. Experiments verify the applicability of the proposed method. The moving platform of the robot successfully reached the target position within 1.5 seconds.