AstroForge wants an AI model to fly its next spacecraft with no uplink
Solo flies in shadow mode on DeepSpace-2 this year, then takes control of Autonomy-1 on Stoke Space's first Nova launch in 2027. The CEO says he may not fly a radio that can receive from Earth.
Shadow mode first, full control second. A spacecraft is a hard place to find out a model was wrong.
AstroForge, a spacecraft company best known for its asteroid missions, announced on September 22, 2026 that an in-house transformer-based model called Solo will control its Autonomy-1 spacecraft in 2027. TechCrunch's Tim Fernholz broke the story. AstroForge's own post says the company does not intend to send the spacecraft a single command after it separates from the rocket.
Autonomy-1 is not an asteroid mission. It carries NASA Goddard's COMPASS payload, which collects space weather and solar data. It launches on the first flight of Stoke Space's Nova Pathfinder, which Stoke targets for early 2027 from Cape Canaveral. Before that, Solo flies in shadow mode on DeepSpace-2, AstroForge's asteroid-bound craft, which rides with Intuitive Machines' third lunar mission by the end of 2026.
What Solo is, and what it is not
Solo does not replace the flight software. AstroForge describes it as an intelligence layer that sits above the spacecraft's deterministic systems. Head of flight software Armand Awad described three parts to TechCrunch: conventional control algorithms, models of subsystems like power and navigation, and an intelligence layer trained on data from about 2,500 sensors. In shadow mode, Solo runs onboard and makes decisions, but the spacecraft does not act on them.
The motive is cost and failure. Ground operations are about a third of mission cost, Runtimewire reports, and AstroForge's 2025 Odin mission lost its early chances to communication problems. CEO Matt Gialich told TechCrunch he does not plan to fly radios that can receive from Earth, though he also said that could change before launch. The spacecraft still sends data down. Stoke's Nova has never flown, so a rocket delay would move Autonomy-1 too.
Why a build studio cares
Shadow mode is the right order of operations, and it is the same safe pattern for adding AI to ordinary software: run the model beside the existing logic, log what it would have done, compare, and only then let it act. What makes Autonomy-1 unusual is the next step. Removing the uplink removes the human override, and in most software the override is the cheapest safety feature there is. The real test of Solo is not the 2027 flight. It is what AstroForge publishes from DeepSpace-2's shadow logs: how often Solo disagreed with the flight software, and which one was right.
Next step: read TechCrunch's report and AstroForge's announcement. If you are adding an AI decision step to software that already works and there is no shadow mode in the plan, write to us at hello@gattyworks.com.