AstroForge puts an AI model in command of its next spacecraft

NASA missions build in heavy redundancy and automation, but still rely on large teams of flight controllers on the ground — the OSIRIS-REx mission, which rendezvoused with an asteroid in 2018, had 100 operators on each eight-hour shift. Startups don’t have those resources, so AstroForge, which is developing technology to mine asteroids, is turning to AI instead. The company has built an autonomous control stack called “Solo,” a transformer-based model made in-house, and plans to fly its first autonomous spacecraft in 2027 on the first rocket launched by Stoke Space, a mission backed by NASA that is expected to gather scientific data about the sun. Most spacecraft autonomy today still depends on traditional control algorithms because of concerns about the reliability of neural networks; the first use of a neural network to control a satellite’s positioning in orbit happened only last year.

AstroForge, founded in 2022 and backed by $56 million in venture funding, has launched two prototype spacecraft, both of which suffered anomalies that prevented them from completing most of their mission objectives. In 2025, its Odin spacecraft was sent into deep space, but the company struggled to communicate with it and ultimately couldn’t regain control — a problem worsened by the limited number of antennas on Earth big enough to transmit to spacecraft hundreds of thousands of miles away, and the narrow windows of time available to do so. “Would that have been recoverable with all the data on the spacecraft? I don’t know, but I can tell you nothing onboard tried it, and I would love something onboard to try if the spacecraft is unrecoverable at launch,” said Matthew Gialich, AstroForge’s co-founder and CEO.

Rather than build its own ground network — which Gialich estimated would cost around $200 million for five dishes worldwide plus operations — AstroForge chose to build more intelligence into the spacecraft itself. Armand Awad, the company’s head of flight software, said the team decided to leverage advances in transformer models driven by frontier AI labs, building a stack that combines traditional control algorithms, models trained on test data for specific subsystems like power generation or navigation, and an overall intelligence layer trained on roughly 2,500 sensors on the spacecraft. Gialich stressed the effort is narrowly scoped: the goal is “constrained autonomy at a very low sensor input,” not general-purpose spacecraft intelligence. In theory, the onboard agent would handle tasks like anomaly resolution — for instance, recognizing it has lost track of its position, correlating a power anomaly with a star-tracker issue, and fixing the problem, likely by power-cycling the affected system.

The company’s third vehicle, DeepSpace-2, is set to launch alongside Intuitive Machines’ third moon mission, expected by the end of 2026. Solo will fly aboard in “shadow mode” so engineers can test it before the Autonomy-1 mission. Gialich said he currently plans to fly Autonomy-1 without radios that can receive signals from Earth, though he acknowledged the team may talk him into adding them before launch.

Why it matters

AstroForge’s bet on onboard AI autonomy, rather than ground control, could reshape how resource-constrained space startups operate missions that larger, better-funded agencies handle with armies of human controllers.