Europe · Onboard AI research and flight demonstration
Edge SpAIce onboard AI research
This project focuses on onboard AI research and flight demonstration.
Edge SpAIce is an Agenium Space-led EU research program that demonstrated a compact marine-litter detection model on EnduroSat's BALKAN 1. CORDIS reports successful image acquisition followed by onboard processing. A separate ground test ran a small neural network on NanoXplore NG-Ultra using CERN HLS4ML. This does not establish a shared data-center service or flight use of NG-Ultra.
Agenium Space-led consortium · Onboard AI workflow demonstrated
Scope: Enabling research · Named enabling AI software and FPGA research with a reported onboard experiment. Not a dedicated shared data-center platform.
Figures retain their disclosed units and scope. Launched hardware does not establish demonstrated performance; targets may change. Not disclosed means no figure in the reviewed sources.
PROJECT RELATIONSHIPS
Suppliers & project partners
CERN
Research participant
Program organization: Agenium Space
Partnership
Provided open-source HLS4ML technology for hardware adaptation, including separate NG-Ultra ground tests.
Research consortium role. NG-Ultra testing is not established as flight hardware.
Announced Publication date not supplied · Evidence checked 28 Oct 2026
End-of-program reporting after November 2026. A proposed ALMAGEST application is not an awarded follow-on.
Manual review 28 Oct 2026. Compared the complete official CORDIS results. BALKAN 1 onboard processing is reported as demonstrated. NG-Ultra testing is ground-only. Optional undisclosed specifications do not invalidate the supported results. Primary body last observed not available. Source monitoring: unavailable. Source collection and substantive profile review are separate checks. Plans and dates may change.
The EU reports that Edge SpAIce demonstrated image acquisition followed by onboard marine-litter detection on EnduroSat BALKAN 1. Agenium Space compressed the model, NTUA trained it and CERN supported hardware adaptation. A separate ground experiment ran a neural network on NanoXplore NG-Ultra.