Hand-eye Calibration for Robonaut

NASA's Human Space Flight program depends heavily on spacewalks performed by human astronauts. These Extra-Vehicular Activities (EVAs) are risky, expensive, and complex. In collaboration with the Defense Advanced Research Projects Agency (DARPA), NASA is developing a robotic astronaut's assistant called Robonaut that can help reduce human EVA time and workload. This project designed and implemented a hand-eye calibration scheme for Robonaut, Unit A. The intent of this calibration scheme is to improve hand-eye coordination of the robot. The basic approach is to use kinematic and stereo vision measurements, namely the joint angles self-reported by the right arm and 3-D positions of a calibration fixture as measured by vision to estimate the transformation from Robonaut's base coordinate system to its hand coordinate system and to its vision coordinate system. Two methods of gathering data sets have been developed, along with software to support each. In the first, the system observes the robotic arm and neck angles, and measures the 3-D position of a calibration fixture using Robonaut's stereo cameras, and logs these data. In the second, the system drives the arm and neck through a set of prerecorded configurations, and data are again logged. Two variants of the calibration scheme have been developed. The full calibration scheme is a batch procedure that estimates all relevant kinematic parameters of the arm and neck of the robot. The " daily " calibration scheme estimates only joint offsets for each rotational joint on the arm and neck, which are assumed to change from day to day. The schemes have been designed to be automatic and easy to use so that the robot can be fully recalibrated when needed such as after repair, upgrade, etc, and can be partially recalibrated after each power cycle. The system has been implemented on Robonaut Unit A and has been shown to reduce mismatch between kinematically derived positions and visually derived positions from a mean of 13.75cm using the previous calibration to means of 1.72cm using the full calibration procedure and 1.85cm using the " daily " calibration procedure. This improved calibration has already enabled the robot to more accurately reach for and grasp objects that it sees within its workspace. The system has been used to support an autonomous wrench-grasping experiment and significantly improved the workspace positioning of the hand based on visually derived wrench position estimates.