AstroForge Bets on Onboard AI to Command Its Next Deep Space Vehicle

AstroForge is preparing to put an artificial intelligence system in direct control of a spacecraft, developing an autonomous transformer-based control stack called Solo that would run the vehicle without ground operators. The asteroid mining startup plans to fly its first autonomous spacecraft in 2027, a move that would break with the industry norm of relying on traditional control algorithms and Earth-based flight teams.

The decision follows setbacks with AstroForge's earlier hardware. The company, founded in 2022 and backed by 56 million dollars in venture funding, has launched two prototype spacecraft, both of which suffered anomalies that prevented them from achieving most of their objectives. In 2025, its Odin spacecraft reached deep space but the company struggled to communicate with it, hampered by the limited number of antennas on Earth large enough to reach vehicles hundreds of thousands of miles away and the narrow windows in which to transmit. AstroForge never regained control of Odin, an experience that pushed it to ask whether a spacecraft could solve its own problems onboard.

Solo combines traditional control algorithms, models trained on test data for individual subsystems such as power generation and navigation, and an overall intelligence layer trained on roughly 2,500 sensors in the spacecraft. Armand Awad, AstroForge's head of flight software, said the company chose to leverage transformer model advances from frontier labs. He described the agent recognizing that it had lost track of its position, correlating a power anomaly to a star tracker fault, and fixing it, probably by turning the component off and on.

Chief executive and co-founder Matthew Gialich framed the effort as a cost trade. Building an in-house ground network with five dishes around the world would cost around 200 million dollars, he said, versus removing that need with a model. "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," Gialich said of Odin. He was clear about the limits of the ambition. "I'm making a constrained autonomy at a very low sensor input, following the basic training of a transformer model," he said.

The plan is notable because most spacecraft autonomy still depends on traditional algorithms, given concerns about the unreliability of neural networks. The first use of a neural network to control a satellite's positioning in orbit took place only last year. For comparison, NASA's Osiris-Rex mission, which rendezvoused with an asteroid in 2018, had 100 operators on each eight-hour shift, the kind of resource AstroForge says startups cannot match.

AstroForge's first autonomous mission is set for 2027 on the first rocket launched by Stock Space, will be backed by NASA, and is expected to gather scientific data about the Sun. Before that, the company's third vehicle, DeepSpace-2, is currently set to launch alongside Intuitive Machines' third Moon mission, expected to head for space by the end of 2026. Solo will fly onboard DeepSpace-2 in shadow mode so engineers can test it ahead of the Autonomy-1 mission. Gialich said he does not plan to fly radios capable of receiving from Earth on Autonomy-1. "We have to go all in right now. The team's probably going to talk me into it by the time we fly it. But right now, I'm telling them no radios," he said.