Autonomous Vision-Based Navigation around Asteroids Using Convolutional Neural Networks

Abstract

This paper investigates the efficacy of Convolutional Neural Network (CNN) based methods to navigate autonomously around asteroids. The main contribution of this work is the successful development of a first-of-a-kind pose estimation pipeline, consisting of a CNN-based feature detector and a Perspective-n-Points (PnP) solver to allow accurate, safe, and autonomous distance estimation with respect to a target asteroid. A top-down CNN-based feature detector is developed, consisting of an object and keypoint detection network in sequence, which detects n pre-defined keypoints, designated on the target's 3D model, within the 2D image. The simulated target asteroid is Bennu, a subkilometer asteroid with a spinning top-shape and pronounced equatorial bulge. The networks have been trained and evaluated on synthetic datasets created in this work, consisting of 32,352 images with a variety of poses from 4.5 to 9 km from the asteroid, for different illumination conditions, asteroid orientations, and image corruptions that emulate real sensor artifacts. The pipeline could achieve a mean and median line-of-sight distance estimate of around 42 m and 30 m, respectively, at a confidence level of 90% for the large relative range, while satisfying the accuracy requirement of a maximum of 10% knowledge error for 99.979% of the cases.