结合IMask R-CNN的绳驱机械臂视觉抓取方法研究

绳驱超冗余机械臂具有灵活性强、工作空间大等特点,在航天活动中可替代宇航员进行各种航空作业.以空间飞行器在轨维修为研究背景,模拟其实验环境,设计了一套基于RGB-D的可移动绳驱超冗余机械臂定位抓取系统.首先改进了Mask R-CNN算法,在保证检测精度的同时降低模型尺寸,通过Intel RealSense D435 i采集图像输入到目标检测模型得到目标的类别和位置信息,进一步利用自适应末端位置更新算法递推机械臂的正逆运动学模型,并结合轨迹规划完成目标的三维空间定位和抓取.实验结果表明,改进后的Mask R-CNN算法能在保证精度的情况下有效地降低模型尺寸,抓取系统的逆运动学求解速度快,具有较好的定位精度,能够有效地完成飞行器抓取的任务.

Visual grasping method of rope-drive manipulator using IMask R-CNN

Rope-driven super-redundant manipulator has the characteristics of strong flexibility and large working space, which can be employed for various aviation operations in space activities to replace astronauts. Based on the research background of spacecraft on-orbit maintenance, this paper simulated the corresponding experimental environment, and designed a RGB-D-based mobile rope-driven super-redundant robotic arm positioning and grasping system. First, it improved Mask R-CNN algorithm to reduce the model size and ensure the detection accuracy. Then it input the image collected by Intel RealSense D435i to the target detection model to obtain the target category and position information. Further it used the adaptive end position update algorithm to recursively calculate the forward and inverse kinematics model of the manipulator, and combined the trajectory planning to complete the three-dimensional space positioning and grasping of the target. The experimental results show that the improved Mask R-CNN algorithm can effectively reduce the model size while ensuring accuracy. Moreover, the inverse kinematics of the grasping system is solved quickly with good positioning accuracy, and can effectively complete the aircraft grasp Task.