SPEED+: Next-Generation Dataset for Spacecraft Pose Estimation across\n Domain Gap

Abstract

Autonomous vision-based spaceborne navigation is an enabling technology for\nfuture on-orbit servicing and space logistics missions. While computer vision\nin general has benefited from Machine Learning (ML), training and validating\nspaceborne ML models are extremely challenging due to the impracticality of\nacquiring a large-scale labeled dataset of images of the intended target in the\nspace environment. Existing datasets, such as Spacecraft PosE Estimation\nDataset (SPEED), have so far mostly relied on synthetic images for both\ntraining and validation, which are easy to mass-produce but fail to resemble\nthe visual features and illumination variability inherent to the target\nspaceborne images. In order to bridge the gap between the current practices and\nthe intended applications in future space missions, this paper introduces\nSPEED+: the next generation spacecraft pose estimation dataset with specific\nemphasis on domain gap. In addition to 60,000 synthetic images for training,\nSPEED+ includes 9,531 hardware-in-the-loop images of a spacecraft mockup model\ncaptured from the Testbed for Rendezvous and Optical Navigation (TRON)\nfacility. TRON is a first-of-a-kind robotic testbed capable of capturing an\narbitrary number of target images with accurate and maximally diverse pose\nlabels and high-fidelity spaceborne illumination conditions. SPEED+ is used in\nthe second international Satellite Pose Estimation Challenge co-hosted by SLAB\nand the Advanced Concepts Team of the European Space Agency to evaluate and\ncompare the robustness of spaceborne ML models trained on synthetic images.\n