A Robot Manipulator Grasping Method Based on Improved YOLOX

The overlap and coverage of objects can affect the grasping success rate for robot manipulator grasping in multi-object scenarios. We propose an enhanced grasping algorithm based on YOLOX-that can predict the bounding box with a smaller aspect ratio, thereby more accurate spatial location. Due to the limits of sensor's environment and physical factors, the depth map will lose some depth values. We propose a depth value repair algorithm based on the FMM algorithm, through which the lost depth values in the grasping region can be repaired. In pose estimation, we use the aspect ratio of the bounding box to determine the rotation angle of the robot manipulator jaws. We use a six-axis robot manipulator combined with a depth camera to achieve object grasping in multi-object scenes. The experimental results show that the enhanced grasping algorithm makes the grasping area prediction more accurate, and the distance between the object and the camera is obtained more accurately.