AstroForge puts AI in command of its next spacecraft with a small transformer model

TechCrunch reported on September 22 that AstroForge is putting AI in command of its next spacecraft: Autonomy-1 will have a small, transformer-based AI model taking charge of the space probe.

The emphasis in that description is on small. The model is not presented as a frontier system but as something that can run under a spacecraft's hardware constraints while still directing the probe.

Autonomy in space is driven by distance. Communication with deep-space probes carries delay, so waiting for instructions from the ground is not always an option when something goes wrong, which is why onboard judgement has long been treated as a way to reduce mission risk.

Handing the command role to a model differs from the older pattern of planning on the ground and executing onboard. It requires inference to run within tight compute, power and environmental limits, which is exactly the problem that small models and compression techniques are built to address.

Transformers are not new to the space world. Over the past few years the architecture has become the default across language, vision and robotics, and shrinking it to fit onboard compute has become a shared engineering goal for AI and aerospace teams.

The report does not disclose the model's size, how it was trained, who supplied it, or what the mission timeline looks like. What can be confirmed so far is the architectural choice itself, not how far it will be tested.

The value of the story lies in the setting. Moving AI from data centres into space forces a re-examination of reliability, interpretability and fallback behaviour: a bad decision by an onboard model costs far more than a failed web request.

What to watch next is how Autonomy-1 performs on its actual mission, and whether AstroForge shares details about the model and the limits of its autonomy. If the approach holds up, onboard AI could move from experiment to default option for more deep-space missions.