AstroForge to fly AI-controlled spacecraft in 2027
AstroForge is developing an autonomous control stack called Solo, a transformer-based AI model, to command its next spacecraft. The mission, launching on Stoke Space's first rocket in 2027, will gather solar data and is backed by NASA. This follows two previous prototype failures, prompting the company to rely on onboard intelligence to solve problems without ground control.
AstroForge's path to autonomy follows two failed prototypes, including the Odin spacecraft that went silent in deep space due to limited ground antenna access. A traditional ground network with five dishes would cost roughly $200 million, versus the in-house Solo model trained on about 2,500 sensors. The stack pairs conventional control algorithms with transformer layers for subsystems like power and navigation.
Before the 2027 Autonomy-1 mission, Solo will ride aboard DeepSpace-2 in shadow mode, letting engineers validate its decisions without relinquishing control. That vehicle launches with Intuitive Machines' lunar mission in late 2026. Gialich intends to fly Autonomy-1 without receiving radios, forcing full independence, though he concedes his team may talk him out of it.
If AstroForge succeeds, autonomous spacecraft could reduce mission costs and open deep-space exploration to startups that lack NASA-scale ground teams. This may accelerate asteroid mining and scientific missions, but it also raises questions about accountability when AI makes irreversible decisions millions of miles away. Society could benefit from cheaper space access, yet the technology's reliability remains unproven, and a failure could set back public confidence in AI-controlled critical systems.