Development of human activities in the space domain has significantly evolved over the past decade, with a growing number of emerging space-faring nations and commercial actors gaining access to the operational environment. The multiplication and diversification of space activities has brought about a stronger need for assessment of space domain awareness (SDA). 

Most of SDA is presently performed through radar detection and tracking or, in the optical domain, by light-curves analysis or laser ranging. These various approaches provide limited information on the resident space objects (RSO) characteristics. In particular, it is difficult or impossible to retrieve the satellite attitude, shape and other relevant information to assess the nature and situation of RSOs. Ground based high resolution imaging systems based on large telescopes and adaptive optics can represent a game changer, by providing high resolution image sequences of resident space objects. These image sequences open the door to new strategies for image processing and estimation of characteristics of RSOs.

ONERA has been working on these approaches for several years and has committed to the development of the largest European ground-based telescope (2.5 m) dedicated to satellite observation, the PROVIDENCE project (optics research platform, vector of innovation for defense on the control and understanding of the environment and characterization of objects in space). First light is foreseen end 2028. This PhD is proposed within this framework and aims at developing image processing strategies to estimate RSOs’ attitude, and if possible RSO’s 3D shape, based on high spatial resolution image sequences similar to what the Providence system shall provide.

The objective is to investigate and combine multiple information extraction methods—such as light curve analysis, silhouette detection, and geometric feature identification—in order to densely capture key image information and mitigate ambiguities. The fusion of these complementary approaches is expected to enhance robustness and reliability. The extracted features will then be used as inputs to Bayesian filters capable of handling a wide range of possible solutions and dealing with severe uncertainties and ambiguities.

The vision-based satellite attitude estimation problem is characterized by a non-injective measurement equation. This causes measurement ambiguities, and in fine state multimodality (several attitudes may correspond to a single satellite observed image). Particle filters are well-known Monte-Carlo methods that were developed to tackle Bayesian inference problems in presence of strongly nonlinear measurement and dynamical models. They proved to outperform classic Kalman-like methods in various applications where measurements are ambiguous (e.g. aerial or underwater robotics). More recently, particle filter methods were adapted to non-Euclidean manifolds such as Riemannian manifolds or Lie groups for aerial navigation involving 3D attitude estimation [1].

However, these approaches fail when the measurement model is not accurately known. The model must then be identified (learnt) online or offline from prior observations. A common way to train a surrogate model is constructing a Gaussian Process from a priori data (for vision-based satellite attitude determination, see [2][3]). An online approach was also introduced to approach the measurement model in a Hilbert space for terrestrial robotics pose estimation [4]. This kind of approach can be taken as a starter for the proposed thesis when applied to the adaptive optics-based attitude estimation problem.

The developed solutions will be tested and evaluated using SIRIUS, an ONERA rendering engine that can simulate in details the hyperspectral images at high spatial and spectral resolution of an RSO, based on its 3D models and surface materials, including the process of image formation with adaptive optics correction. Validation on real images will also be considered, either on a smaller telescope or with PROVIDENCE’s first images by the end of the thesis. 

The developped algorithms will also be applicable to onboard applications such as space rendez-vous and on-orbit servicing.

[1] Chahbazian et al. (2021, November). Laplace particle filter on lie groups applied to angles-only navigation. In 2021 IEEE 24th International Conference on Information Fusion (FUSION) (pp. 1-8). IEEE.

[2] Liu et al. (2024). Attitude takeover control for noncooperative space targets based on gaussian processes with online model learning. IEEE Transactions on Aerospace and Electronic Systems, 60(3), 3050-3066.

[3] Hara et al. (2025). Attitude estimation from photometric data using Gaussian process regression. Journal of Space Safety Engineering.

[4] Kok et al. (2018, July). Scalable magnetic field SLAM in 3D using Gaussian process maps. In 2018 21st international conference on information fusion (FUSION) (pp. 1353-1360). IEEE.

=================

For more Information about the topics and the co-financial partner (found by the lab!); contact Directeur de thèse - cyril.petit@onera.fr

Then, prepare a resume, a recent transcript and a reference letter from your M2 supervisor/ engineering school director and you will be ready to apply online  before March 13th, 2026 Midnight Paris time!