Development and In-Orbit Operation of Deep Learning Attitude Sensor

Abstract

This study presents in-orbit demonstrations of a deep learning attitude sensor (DLAS) aboard the Japan Aerospace Exploration Agency (JAXA) small-satellite, Rapid Innovative Payload Demonstration Satellite 1. The DLAS is a spaceborne component that consists of star trackers, Earth cameras, and onboard computers. We propose a novel attitude determination method that uses a small camera and image recognition algorithm on board the DLAS to achieve three-axis attitude determination with minimal weight and volume. The DLAS mission further includes a demonstration of a low-cost high-performance star tracker that integrates consumer products. The development of the DLAS began in 2016 when it was adopted as the theme of the JAXA Innovative Satellite Technology Demonstration program. It was launched into a sun-synchronous orbit at an altitude of 500 km in January 2019 for approximately one year. This paper describes the results obtained from the DLAS operation and its future applications. The DLAS demonstrated an Earth camera image identification accuracy of more than 70%. Moreover, the obtained attitude determination accuracy was approximately 2.8 deg for the Earth camera and approximately 14 arcs for the star trackers. The technologies developed in the DLAS have been implemented in a subsequent microsatellite (HIBARI) of the Tokyo Institute of Technology.