Target Recognition and Optimal Grasping Based on Deep Learning

The difficulty of robot in three-dimensional target recognition and optimal grasping is the complex background environment and the irregular shape of the target object. The robot is required to identify the best grasping pose of the target like a human while identifying different three-dimensional objects posture. A deep learning method based on the cascaded faster-rcnn model to identify the target object and its optimal grasping posture is proposed in this paper. First, the improved faster-rcnn model is used to identify the object and determine its approximate position. Then, the other faster-rcnn model is used to find the optimal grasping pose of the target, and complete robot's optimal grasp. Experiments show that the method can quickly and accurately find the target object and determine its optimal grasping pose.