ECLIPSE: A Spacecraft Rendezvous Trajectories Dataset with Controlled In-Orbit Lighting Conditions

Nidhal Eddine Chenni Arunkumar Rathinam Abid Ali Djamila Aouada
Interdisciplinary Centre for Security, Reliability and Trust (SnT), University of Luxembourg
NeurIPS 2026, Track on Evaluations and Datasets
Paper Dataset Code Weights
Synthetic renders of the Proba-2 satellite next to real laboratory captures of the same mockup under lighting conditions L1 to L4, which become progressively darker.
Add Figure 1 at static/images/teaser.png

Two axes of domain shift in ECLIPSE: the sim-to-real gap between synthetic and real images, and the illumination gap within the real domain, from frontal lighting (L1) to near-eclipse backlighting (L4).

Abstract

Deep learning has become the dominant paradigm for vision-based spacecraft pose estimation, where models are predominantly trained on synthetic data due to the prohibitive cost of acquiring real in-orbit imagery. When deployed, these models suffer significant performance degradation due to the sim-to-real domain gap, which existing benchmarks address by pairing synthetic training data with Hardware-in-the-Loop (HIL) real imagery. Yet a critical factor remains consistently overlooked: in real orbital scenarios, illumination varies with the spacecraft's position along its orbit, inducing severe appearance shifts that existing benchmarks neither control nor study.

We introduce ECLIPSE, the first benchmark designed to isolate illumination as an independent experimental variable. ECLIPSE pairs a large-scale photorealistic synthetic training set with a fully labeled HIL evaluation set of 14 approach trajectories, each replayed identically under 4 controlled lighting conditions, enabling rigorous evaluation of illumination-robust and domain-adaptation methods. Through this controlled design, our experiments confirm that illumination is a major driver of performance degradation in spacecraft pose estimation, both within the real domain and across the sim-to-real gap.

Four Controlled Lighting Conditions

Each real trajectory is recorded four times with the same motion, once per lamp position, so only the lighting changes.

Top view of the setup: the lamp moves around the target from the camera side (L1) to behind the target (L4). L1 L2 L3 L4 camera
Real capture of the mockup under lighting condition L1

Data and Code

Dataset

Synthetic and real splits with pose annotations.

Hugging Face

Code

Baseline training and evaluation code, and rendering utilities for extra labels.

GitLab

Model Weights

Trained weights for the baseline models in the paper.

Download

Acknowledgements

This work was supported by the Luxembourg National Research Fund (FNR) under the project reference BRIDGES/2025/IS/19855391/U-ADAPT, and by Infinite Orbits. Experiments were performed on the Luxembourg national supercomputer MeluXina at LuxProvide.

Luxembourg National Research Fund (FNR) Infinite Orbits

BibTeX

@inproceedings{chenni2026eclipse,
  title     = {{ECLIPSE}: A Spacecraft Rendezvous Trajectories Dataset
               with Controlled In-Orbit Lighting Conditions},
  author    = {Chenni, Nidhal Eddine and Rathinam, Arunkumar and
               Ali, Abid and Aouada, Djamila},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS),
               Track on Evaluations and Datasets},
  year      = {2026}
}