The ISOLDE Space Demonstrator: A Platform for AI Applications on Satellites

Deploying AI-based functionalities to satellites enhances spacecraft autonomy but presents significant challenges for the design of both hardware and software platforms. Advantages of onboard processing include secure and private computing, the reduction of data uplink/downlink requirements, autonomous anomaly detection and recovery, and in general autonomous spacecraft operation. However, edge inference is computationally intensive and has to be performed onboard while sharing resources with other traditional applications, like avionics, Attitude Orbit Control, and signal/telemetry processing. To enable secure and energy-efficient AI-processing capabilities without decreasing the performance of traditional applications, it is necessary to take an holistic approach, jointly optimizing the hardware and software components. To achieve these goals, within the ISOLDE project, a space demonstrator is developed. In this work, the general architecture and main components of the demonstrator are presented, with a particular focus on the AI applications under development. One of the main objectives of the project is the achievement of technological sovereignty, with open hardware and software solutions. Within this project, a new Linux-capable platform based on RISC-V processors and accelerators is developed, enabling the execution of traditional and AI-based workloads on satellites. The project will target hardware and software codesign to optimize the performance of the platform, prioritizing energy efficient, secure and error resilient computing, and will develop a set of applications to demonstrate the capabilities of the platform.