6D pose estimation and grasping based on Deep learning with MBM

摘要

Aiming at the problem of low recognition rate and poor robustness of traditional robotic manipulation visual recognition algorithm, an algorithm based on Minimum Bounding Model (MBM) is introduced based on existing deep learning 3D vision research in this paper to accurately detect and identify the target. This method is based on the improved YOLO algorithm and is trained with data generated by a new data generation method. Through a single RGB image, the object can be directly identified and the 6D pose information can be estimated simultaneously. Based on the above process, the path planning algorithm can be used to capture the object. Experimental tests prove that the method can accurately identify objects and perform pose estimation. The grasping experiment is carried out on the Co602a manipulator. The results show the effectiveness of the method.