AstroForge puts Solo AI in command of its next spacecraft
Asteroid-mining startup AstroForge is developing an in-house AI model called Solo to autonomously pilot its upcoming spacecraft, bypassing the need for expensive ground control teams.
AstroForge, an asteroid-mining startup founded in 2022 with $56 million in venture funding, has built an in-house transformer-based control stack named Solo. The AI agent is designed to autonomously operate spacecraft, resolving anomalies without human intervention. Solo will first fly in shadow mode aboard the DeepSpace-2 vehicle, scheduled to launch alongside Intuitive Machines' third lunar mission by the end of 2026. If successful, the model will take full control of the Autonomy-1 mission in 2027, launching on Stoke Space's inaugural rocket for a NASA-backed solar research mission.
The shift to AI-driven autonomy comes after AstroForge faced severe communication issues with its Odin spacecraft in 2025. Because deep-space communication requires massive, scarce Earth-based antennas, the startup struggled to maintain contact. According to CEO Matthew Gialich, building a proprietary global ground network of five dishes would cost roughly $200 million. Instead of pursuing that expensive infrastructure or employing massive flight control teams—such as the 100-operator shifts NASA used for its OSIRIS-REx mission—AstroForge is betting on onboard intelligence to solve unexpected flight issues.
Developed by flight software head Armand Awad, the Solo architecture merges traditional control algorithms with specialized models trained on subsystem test data. An overall intelligence layer is trained on data from approximately 2,500 sensors across the spacecraft. The system is designed to identify and resolve complex issues, such as correlating power anomalies with star tracker failures. Gialich is so confident in this approach that he plans to fly the 2027 Autonomy-1 mission without any receiving radios, forcing the spacecraft to rely entirely on Solo.
For space industry practitioners, this represents a radical departure from traditional aerospace engineering, which historically avoids neural networks for critical guidance systems due to unpredictability concerns. If AstroForge succeeds, it could prove that constrained transformer models can safely manage deep-space operations, drastically lowering the barrier to entry for private interplanetary exploration.
This is our own summary of reporting by TechCrunch AI
