AstroForge entrusts the piloting of its next ship to an AI named Solo, initially in observer mode

The system, built on a Transformer architecture, reads about 2,500 sensors to diagnose the status of the spacecraft and trigger maneuvers, without going through ground crews.

On September 22, 2026, AstroForge detailed Solo, the autonomy system that is to take control of its next spacecraft. The architecture is divided into three layers: traditional control algorithms as close as possible to the hardware, specialized models by subsystem — power generation, navigation — and then a decision-making layer built on a Transformer architecture, which aggregates data from about 2,500 sensors to establish a diagnosis and trigger the appropriate response.

The stated objective is to reduce dependence on ground crews and long-distance communication networks, two well-known bottlenecks in space exploration. However, Solo will not take over immediately: it will first run in observer mode aboard DeepSpace-2, which is expected to be launched at the end of 2026 as part of an Intuitive Machines lunar mission.

If the test goes well, the company is aiming for a fully autonomous mission in 2027, with the support of NASA. The context helps: the first use of a neural network to control the positioning of a satellite in orbit dates back only to the previous year.