Feasibility of a COTS GPU-Based On-board AI Platform for Robotic Space Missions

Kurzfassung

The feasibility of deploying a high-performance COTS GPU platform for robotic space missions is assessed, focusing on NVIDIA Jetson modules integrated into the ScOSA architecture. The design couples the Jetson with a radiation-hardened Co-CPU for active fault detection, error correction, and controlled shutdown during high-radiation events, while exploiting the GPU's AI capabilities for real-time perception and manipulation tasks. Radiation simulations for a representative LEO demonstrate that aluminum shielding reduces the TID to well below the Jetson's tested tolerance. SEEs remain a concern and are addressed through system-level mitigation. Thermal modeling shows that a PCM buffer significantly extends operational run-time compared to passive heatsinking alone, enabling longer AI inference periods without active cooling.