Optimizing Multi-Task Learning for Accurate Spacecraft Pose Estimation

TL;DR

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.

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.

Figures (5)

• Figure 1: Two samples from our dataset. Each row corresponds to a sample. The first column contains the generated capture. The second column shows the bounding box. The third column displays the keypoints' heatmap. The fourth column contains the segmentation masks.
• Figure 2: Simplified visualization of the proposed CNN MTL architecture. P, H, B, and S respectively represent the heads for direct pose estimation, keypoints' heatmaps prediction, bounding box prediction, and segmentation tasks.
• Figure 3: Percentual change in SPEED score from direct pose estimation compared to the single task network P performance. A positive change means a reduction in the score (the lower the better).
• Figure 4: Percentual change in SPEED score from indirect pose estimation compared to the single task network H performance. A positive change means a reduction in the score (the lower the better).
• Figure 5: Inference results on a test sample. The predicted bounding box, heatmaps and segmentation are highlighted in the bottom row, while the relative ground truths are in the upper row.