I have a KUKA iiwa 14 and a realsense D435i camera. I calibrated the system using a hand-to-eye configuration in Moveit! Hand-eye calibration package. But I would now like to test the accuracy of the calibration. Any idea how this can be done? Do I need to manually feed in different known points in space and command the robot to move to it? How could I calculate the error between desired location and where the robot reached in this case?
Not a very scientific method, but a way I've been doing it is when the calibration is done, the algorithm usually gives out the calibration frame as well, which is a transformation from the robot base_link to the coordinate point where the calibration board starts (depending on if you're using checkers, circles or other it might be the crossing point or the larger circle).
You can use this calibration frame and make it as a frame in your .xacro setup, which you can then see as a tf_transform in, for example, RViz. If you have a 3D model of the system with the calibration board in RViz, you can publish the pointcloud of your D435i camera and compare if the frame you made is in the place the camera sees the calibration board.
But I would suggest checking out @Markus's answer on the Hand-Eye Calibration Residuals for a more scientific method.
I recommend you check the information "howto" here: enter link description here