Advanced Synthetic Data and Pose Estimation Framework for Enhanced Autonomous Spacecraft Operations

We aim to bridge the current domain gap in visual-based deep learning solution in space activities by:

1. Generating high-fidelity Synthetic Data Leveraging state-of-the-art path-tracing and real-time 3D tools to create photorealistic synthetic data sets. This data will closely mirror real-world space environments, substantially improving the training and performance of deep learning models used in space applications.
2. Establishing a robust Simulation Testbed Introducing a versatile, real-time simulation environment integrated with an n-body simulation and a comprehensive software API. This testbed will enable the formulation and safe testing of various space activity scenarios, allowing for risk-free evaluation and refinement of visual-based deep learning models.
3. Innovating 6-Degrees-of-Freedom Pose Estimation Advancing current 6DoF pose estimation techniques by incorporating more robust keypoint detection, probabilistic Perspective-n-Point (PnP) solving, and solution deployment on space-grade edge devices like AMD Versal AI Edge. This innovation aims to improve automatic visual navigation in next-generation satellites, enhancing autonomous proximity operations' safety, efficiency, and success.

Our goal is to address and overcome the challenges of domain-specific training limitations and push the boundaries of what is achievable in autonomous space missions, aligning closely with ESA's vision of advancing space technology and exploration.