Generative Maneuver Inference for Non-Cooperative Objects in Cislunar Space During Optical Blackouts Using Diffusion Models

In cislunar space, multi-body gravitational dynamics and limited opportunities for optical observation—due to solar interference and other factors—make it challenging to detect maneuvers of non-cooperative objects based solely on angle-only tracklet measurements. As utilization of the cislunar region grows, advancing Space Situational Awareness technologies to mitigate risks associated with collisions and proximity operations becomes essential. This study proposes a probabilistic method that simultaneously infers both the maneuver profile during periods of optical blackout and the measurement-to-object association for multiple non-cooperative objects in cislunar space. The method utilizes Conditional Diffusion Models and relies solely on a single angle-only tracklet obtained after the blackout period. Simulation results show that the proposed conditional diffusion model successfully recovers the maneuver as a well-resolved multimodal probability distribution. In addition, in scenarios where the measurement-to-object association is unknown, the method is able to correctly infer the correlation between measurements and objects.