Robotic Testbed for Rendezvous and Optical Navigation: Multi-Source\n Calibration and Machine Learning Use Cases

摘要

This work presents the most recent advances of the Robotic Testbed for\nRendezvous and Optical Navigation (TRON) at Stanford University - the first\nrobotic testbed capable of validating machine learning algorithms for\nspaceborne optical navigation. The TRON facility consists of two 6\ndegrees-of-freedom KUKA robot arms and a set of Vicon motion track cameras to\nreconfigure an arbitrary relative pose between a camera and a target mockup\nmodel. The facility includes multiple Earth albedo light boxes and a sun lamp\nto recreate the high-fidelity spaceborne illumination conditions. After the\noverview of the facility, this work details the multi-source calibration\nprocedure which enables the estimation of the relative pose between the object\nand the camera with millimeter-level position and millidegree-level orientation\naccuracies. Finally, a comparative analysis of the synthetic and TRON simulated\nimageries is performed using a Convolutional Neural Network (CNN) pre-trained\non the synthetic images. The result shows a considerable gap in the CNN's\nperformance, suggesting the TRON simulated images can be used to validate the\nrobustness of any machine learning algorithms trained on more easily accessible\nsynthetic imagery from computer graphics.\n