【摘要】 航天用电连接器的有效抓取是完成在轨装配任务的前提，有效抓取点集合受到装配任务的限制。本文提出一种面向在轨装配的电连接器抓取方法，该方法通过机器人在模拟环境中学习抓取经验，在抓取-装配过程中评估抓取位姿与装配任务相关性抓取质量，此外，通过设计抓取点的位置分布初步考虑了电连接器尾端线缆对抓取的影响。利用抓取网络对抓取点云集合进行训练获取抓取编码簿，一方面可以生成有效抓取点，另一方面通过建立的表面接触热图增强有效抓取点的可展示性。同时，提出一种非统一归一化物体坐标空间表示法，在三维空间中建立密集映射关系，将有效抓取点迁移到同类别实例中。仿真结果表明：基于本文提出的方法生成的抓取点集合有94.1%有效用于后续的装配任务。更多还原

【Abstract】 Effective grasping of aerospace electrical connectors is a prerequisite for completing on-orbit assembly tasks,and the set of effective grasping points is limited by the assembly tasks. In this paper,we propose an on-orbit assembly-oriented electrical connector grasping method,which learns the grasping experience from the robot in a simulated environment,and evaluates the grasping quality by detecting the correlation between the grasping position and the assembly task during the grasping-assembly process.In addition,this paper preliminarily takes into account the effect of the electrical connector’s tail-end cable on the grasping through the design of the position distribution of the grasping points. Then a grasping network is used to train the grasping point cloud collection to obtain the grasping codebook,which can generate effective grasping points on the one hand,and enhance the demonstrability of effective grasping points through the established surface contact heat map on the other hand. At the same time,this paper proposes a non-uniform normalised object coordinate space representation to establish a dense mapping relationship in 3D space to migrate the effective grasping points to the same class of instances.Simulation results show that 94.1% of the gripping point sets generated based on the method proposed in this paper are effectively used for subsequent assembly tasks.更多还原