AstroForge is putting AI in command of its next spacecraft
TechCrunch Tim Fernholz
AstroForge is putting an AI control system on its next spacecraft. It wants the ship to fix itself in space, not wait for Earth to save it.
Based on reporting by TechCrunch, Tim Fernholz — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
AstroForge wants to send software where most space companies still send people. The asteroid-mining startup says its next spacecraft will run on an in-house autonomous control stack called Solo, built around a transformer model and designed to handle problems without waiting for Earth.
That is a big swing for a company that has already had to deal with the hard parts of space the old-fashioned way. AstroForge has launched two prototype spacecraft, and both missed most of their mission goals after running into anomalies. In 2025, its Odin craft went deep into space, then the team struggled to communicate with it. Ground antennas that can reach something that far away are scarce, and the windows to use them are small. In the end, AstroForge never got Odin back under control.
That failure changed the company’s thinking. Chief executive Matthew Gialich says the alternative to more onboard intelligence could be building a ground network of its own, which he pegs at around $200 million for five dishes around the world. Instead, AstroForge is leaning into models, with Armand Awad, its head of flight software, saying the company built a stack that combines traditional control systems, models trained for specific subsystems, and a broader intelligence layer trained on about 2,500 sensors in the spacecraft.
The pitch is not general-purpose AI in space. Gialich is explicit about that. He calls it constrained autonomy, with a narrow sensor input and a transformer model trained for a specific job. In theory, that agent could spot an anomaly, connect it to a broken star tracker, and recover by switching a component off and on again. That sounds modest until you remember how much of spaceflight still depends on people on the ground.
AstroForge’s third vehicle, DeepSpace-2, is due to launch with Intuitive Machines’ third moon mission by the end of 2026, and Solo will fly there in shadow mode first. The real test comes with Autonomy-1 in 2027, when AstroForge plans to launch its first fully autonomous spacecraft on Stoke Space’s first rocket, with NASA backing and a science goal tied to the Sun. The company says that flight may not even carry radios for Earth contact. That’s either bravery or the kind of confidence that only comes after getting burned.
My take — AI-written commentary, not fact-checked reporting
This is the right kind of AI in space: narrow, ugly, useful. The industry has spent years worshipping autonomy in the abstract, then quietly paying armies of humans to babysit the hardware anyway. If AstroForge can get a spacecraft to notice it’s confused and untangle itself, that beats another glossy demo of a model answering questions it’ll never be asked in orbit.
Read more about this at: TechCrunch
Related stories
Former SpaceX engineers are building a robotic factory for making steel parts
Ars Technica · 1 month ago ·
33
SpaceX designed an orbital Vera Rubin. Radiation comes next.
The New Stack · 3 weeks ago ·
51
OpenAI’s Astra can do a researcher’s week of work. That’s the problem.
The New Stack · 3 weeks ago ·
18