The rise of megasatellite constellations with advanced sensors generating vast amounts of data is revolutionizing the space industry. These constellations demand new methods for data analysis, storage, processing, and delivery. A transformative solution is to shift data processing tasks to the edge, i.e., to the satellites themselves.

About the Project

SpaceEdge 2 is a collaboration between Unibap and AI Sweden, aimed at refining edge technology for handling terrestrial and orbital data and leveraging this data for the development and training of AI/ML applications. By engaging with AI Sweden's space projects and partner ecosystem, the project explores applications deployable on the SpaceEdge platform to test and validate these applications. This project involves stakeholders with expertise in AI and space technology.

Challenges

The space infrastructure, particularly focused on Earth Observations (EO), generates immense data volumes for training ML models to make real-time decisions on the ground. Understanding model behavior and reliability under various conditions is crucial. Research is needed to compare the performance of ML models and data handling on edge devices tailored for space missions. As satellite constellations become the norm, addressing the challenges of distributed computing, decentralized learning, and model interpretation in this setup is essential.

Expected Outcomes

The primary goal is to become an international leader in edge learning and computing in space. The project aims to demonstrate end-to-end onboard AI processing using national competencies and resources from Unibap, AI Sweden Edge lab, and other contributors in Sweden. Moving computation to the edge and utilizing decentralized learning will:

- Reduce communication delays
- Improve the quality of training data
- Minimize the bandwidth needs for ground-to-satellite communication

This will foster innovation in applied AI and ML, prioritize model explainability, support distributed computing and decentralized learning, and emphasize the need for computation and software infrastructure designed for space.

Status

The current software version enables the management of satellite applications from development through deployment to execution. Hardware testing is underway, and reports will define requirements for next-generation payload hardware design.

SpaceEdge 2 significantly contributes to Agenda 2030, advancing the use of distributed computing and learning algorithms in space, and enhancing knowledge about the latest CPU/GPU solutions and AI/ML accelerators for satellite deployment.