AstroForge is preparing to put a transformer-based AI system in charge of a spacecraft, reducing its dependence on scarce ground antennas and large flight-control teams. The in-house stack, called Solo, combines traditional control software, models for individual subsystems and a coordinating layer trained on data from about 2,500 sensors.
The asteroid-mining startup’s first two prototype spacecraft suffered anomalies that blocked most mission goals. Engineers struggled to communicate with the deep-space Odin vehicle, prompting the company to ask whether onboard software could diagnose and recover from failures without waiting for Earth.
Solo will first fly in “shadow mode” on DeepSpace-2, currently expected to launch with Intuitive Machines’ third lunar mission by the end of 2026. It can then be compared with the real control system without commanding the craft. A 2027 mission called Autonomy-1 is planned as the first operational test and is expected to collect solar science data with NASA backing.
Chief executive Matthew Gialich has discussed flying Autonomy-1 without a radio that receives commands from Earth, though he acknowledged his team may change that plan. The staged shadow test is crucial because neural-network control remains unusual in spacecraft and AstroForge’s earlier missions exposed the cost of limited recovery options.