AstroForge to fly AI-commanded spacecraft in 2027
Original: AstroForge is putting AI in command of its next spacecraft
Why This Matters
AI-controlled deep-space craft could lower mission costs dramatically for commercial space players.
Asteroid mining startup AstroForge has built an in-house transformer-based autonomous control system called 'Solo' for its spacecraft. The $56M-backed company plans to fly the first Solo-equipped mission in 2027 aboard a Stoke Space rocket, backed by NASA, after two prior spacecraft suffered mission-critical anomalies.
AstroForge, founded in 2022 with $56 million in venture funding, is replacing conventional ground-control operations with an onboard AI system called 'Solo' — a transformer-based model built entirely in-house. The decision was driven by hard experience: its 2025 Odin spacecraft was lost in deep space after the company could not establish reliable communication through the limited number of large ground antennas available for deep-space missions.
CEO Matthew Gialich put the economics bluntly: building a proprietary ground antenna network would cost around $200 million. Using an onboard AI model is the alternative. '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.
Head of flight software Armand Awad described Solo's architecture as layered: traditional control algorithms at the base, subsystem-specific models trained on test data for areas like power and navigation, and a top-level intelligence layer trained on roughly 2,500 spacecraft sensors. The 2027 mission, launching on Stoke Space's first rocket and supported by NASA, will gather solar scientific data — and serve as the real-world test of the approach.
Neural-network-based spacecraft control remains rare. The first use of a neural network to manage a satellite's orbital positioning happened only in 2025, making AstroForge's ambitions a significant technical bet.