After recognizing the target pose by visual perception,the robot hand eye calibration is needed to move to the specified pose correctly.To perform manipulation in three-dimensional space,the three-dimensional hand-eye calibration is needed for robot.A three-dimensional hand eye calibration method was proposed for the space object grasping task:the robot workspace was formed by setting the limits of the robot in the x,y and z directions,and a working area was defined for the robot according to work needs;the robot moved in the area with a fixed step and a fixed trajectory route,and coordinate system conversion matrix was obtained through iterative closest points algorithm,it was optimized through the Nelder-Mead method to reduce calibration error.Through experiments,it is concluded that the calibration error is 18.90% less than the traditional baseline method,and the time is reduced by 49.12% compared with the baseline method.