Imagine sending a spacecraft to an asteroid, landing it perfectly, and having everything go flawlessly. If only it were that simple. NASA tackles solar system exploration by loading its vehicles with redundancy and automation, yet a significant portion of the work still falls to teams of flight controllers on Earth who stay in constant communication with the craft. Take the Osiris-Rex mission, which met up with an asteroid back in 2018—it required 100 operators working every eight-hour shift. Startups simply can't match that kind of manpower, so AstroForge, a company building asteroid mining technology, is betting on AI instead.

EMBED_PLACEHOLDER_0

The firm has created its own autonomous control system for spacecraft, dubbed "Solo," which is a transformer-based model developed internally. AstroForge intends to launch its first autonomous spacecraft in 2027, riding on the debut rocket from Stock Space. NASA will back that mission, which is also expected to collect scientific information about the Sun. Pulling that off would be a major feat, since most spacecraft autonomy today still relies on conventional control algorithms—largely because neural networks are seen as too unreliable. In fact, it was only last year that a neural network was first used to manage a satellite's positioning in orbit.

Founded in 2022 and armed with $56 million in venture funding, AstroForge has already sent up two prototype spacecraft, and both ran into anomalies that kept them from completing most of their objectives. In 2025, the company's Odin spacecraft launched into deep space, but staying in touch with it proved difficult. Earth has only a handful of antennas powerful enough to reach spacecraft hundreds of thousands of miles away, and the available communication windows are brief. In the end, AstroForge never managed to control Odin, but that setback pushed the company to explore other options: what if the spacecraft itself carried enough intelligence to work through its own problems?

"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. "The trade for me is: Do I go build my own ground network, which is going to cost [around] $200 million to put up five dishes around the world and then do operations on it, or do I try to remove it with a model."

Armand Awad, who leads flight software at AstroForge, explained that the company chose to build on the breakthroughs in transformer models coming from the frontier labs. The resulting stack combines traditional control algorithms, models trained on test data for particular subsystems such as power generation or navigation, and a broader intelligence layer trained on roughly 2,500 sensors within the spacecraft. "I'm not saying I'm going to make general spacecraft autonomy or general autonomy for the world," Gialich said. "I'm making a constrained autonomy at a very low sensor input, following the basic training of a transformer model."

In principle, the agent controlling the spacecraft would handle things like resolving anomalies. Awad pictures it noticing that it has lost its position in space, connecting a power anomaly to problems with its star tracker, and then fixing the entire issue—"probably turning it on and off in this case."

The company's third vehicle, DeepSpace-2, is slated to launch alongside Intuitive Machines' third moon mission, which is expected to depart for space by the end of 2026. Solo will ride along on that vehicle, operating in "shadow mode," giving AstroForge's engineers a chance to test it thoroughly before the Autonomy-1 mission. So will this really be an AI agent running a spacecraft independently? "I don't plan on flying radios that can receive from Earth on Autonomy-1," Gialich said. "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."