DreamSat-Pose: Spacecraft Pose Estimation from Single-View 3D Reconstructions and Learned 2D-3D Feature Matching

Josiane Uwumukiza, Jocelyn Zhao, Giovanni Lavezzi, Giacomo Battaglia, Paolo Panicucci, Minduli C. Wijayatunga, Victor Rodriguez-Fernandez, and Richard Linares

Paper Code

Method Overview

Overview of the DreamSat-Pose framework — DreamSat-Pose reconstructs a 3D model from a single RGB image, extracts 2D and 3D features, learns dense correspondences, and estimates the spacecraft pose using a Perspective-n-Point solver.

Abstract

6-DoF pose estimation is a critical task in autonomous rendezvous and proximity operations. In the case of an unknown target, this task becomes challenging as it shall be paired with the reconstruction of the target shape model. In this article, we propose a novel framework for single-shot shape and pose estimation of unknown spacecraft objects. Given a single image, we first reconstruct a 3D shape model of the target, then estimate the relative six-degrees-of-freedom pose by learning dense 2D-3D correspondences. The image features are extracted using a frozen DINOv3 vision transformer, while the geometric features are computed from the reconstructed point cloud using a trainable dynamic graph convolutional neural network encoder. A dual-stream transformer matcher refines descriptors through alternating self- and cross-attention, producing soft correspondences that are passed to a Perspective-n-Point solver for pose recovery. We evaluate the method on the SPE3R dataset and consider FoundationPose as a representative baseline for current state-of-the-art capabilities. Results show reliable pose estimates, achieving a mean pointing error of 0.157 degrees using only a single image and reconstructed geometry, and demonstrating strong generalization to unseen spacecraft.

Qualitative Examples

Each example shows the original spacecraft image and the estimated pose compared with the reference pose.

Aquarius

CYGNSS Solo 39

EO-1

EO-1 estimated pose compared with the reference pose — Estimated pose and reference pose

Galileo

Herschel

Loral 1300

SMART-1

BibTeX

@misc{uwumukiza2026dreamsatpose,
  title = {DreamSat-Pose: Spacecraft Pose Estimation from Single-View
           3D Reconstructions and Learned 2D-3D Feature Matching},
  author = {Uwumukiza, Josiane and
            Zhao, Jocelyn and
            Lavezzi, Giovanni and
            Battaglia, Giacomo and
            Panicucci, Paolo and
            Wijayatunga, Minduli C. and
            Rodriguez-Fernandez, Victor and
            Linares, Richard},
  year = {2026},
  eprint = {2607.13449},
  archivePrefix = {arXiv},
  primaryClass = {cs.CV}
}