基于最小尺寸点模型的6D位姿估计与机械臂抓取

6D Pose estimation and robotic arm grabbing based on minimum size points model

针对传统机械臂视觉识别算法识别率低、鲁棒性差、运行时间长等问题,进行了基于深度学习3D目标检测的研究,提出了一种基于最小尺寸点模型的目标检测与姿态估计的抓取方法。该方法基于改进的YOLO算法,同时以所提出的数据集建立方法构建数据集进行训练,通过处理单张RGB图,即可对目标物体进行识别并估计其6D位姿信息,在此基础上再结合路径规划算法对目标物体进行抓取。通过仿真实验证明了该方法能准确地对物体进行分类与位姿估计。在Co602a机械臂下进行了抓取实验,结果证明了该方法的有效性。

Aiming at the problems of low recognition rate,poor robustness and long running time of traditional ro-botic arm visual recognition algorithms,a grab method for object detection and pose estimation based on minimum.size points model was proposed.On the basis of the improved YOLO algorithm,the dataset was established with the proposed grab method to train the data set.Through a single RGB image,the object could be directly detected and the 6D pose information could be estimated simultaneously.On this basis,the path planning algorithm was used to capture the object.Simulation experiment proved that the method could accurately classify objects and perform pose estimation.The gripping experiment was carried out on Co602a manipulator,and the results showed the effective-ness of the method.