Robust on-hand robot-3D camera calibration using gradient descent in noisy sensor environments

This paper presents preliminary results on a novel gradient descent-based method for on-hand robot-camera calibration, addressing the challenges posed by sensor noise. The proposed method, grounded in the AX=XB paradigm, effectively resolves the transformation matrix between the 3D camera frame of reference and the robot's flange frame of reference, achieving a calibration accuracy of 15 mm on test data. The experimental setup includes a KUKA KR16 robot base and an IFM O3D313 time-of-flight camera, with a 200x200 mm AprilTag code positioned within the robot's workspace. The method requires the location of the tag in the robot’s workspace and the camera’s position in respect to Robot’s flange to remain fixed while the tag's position in the camera's frame of reference is established through a detection algorithm.