Optimizing Multi-Task Learning for Accurate Spacecraft Pose Estimation

TL;DR

This paper investigates how different task combinations within a multi-task learning framework affect the accuracy of satellite pose estimation from monocular images, highlighting the positive influence of direct and heatmap-based pose tasks.

Contribution

The study systematically evaluates the impact of various task configurations in a modular CNN for satellite pose estimation, revealing which tasks enhance or hinder accuracy.

Findings

Direct and heatmap-based pose tasks mutually improve each other.

Bounding box and segmentation tasks do not significantly contribute and may degrade performance.

A synthetic dataset was created for training and testing the multi-task network.

Abstract

Accurate satellite pose estimation is crucial for autonomous guidance, navigation, and control (GNC) systems in in-orbit servicing (IOS) missions. This paper explores the impact of different tasks within a multi-task learning (MTL) framework for satellite pose estimation using monocular images. By integrating tasks such as direct pose estimation, keypoint prediction, object localization, and segmentation into a single network, the study aims to evaluate the reciprocal influence between tasks by testing different multi-task configurations thanks to the modularity of the convolutional neural network (CNN) used in this work. The trends of mutual bias between the analyzed tasks are found by employing different weighting strategies to further test the robustness of the findings. A synthetic dataset was developed to train and test the MTL network. Results indicate that direct pose estimation and heatmap-based pose estimation positively influence each other in general, while both the bounding box and segmentation tasks do not provide significant contributions and tend to degrade the overall estimation accuracy.