AstroForge leverages AI for autonomous spacecraft control

AstroForge, a startup developing asteroid-mining technology, is pioneering the use of AI to control its spacecraft. The company has developed an autonomous control system called 'Solo,' a transformer-based model, to address the limitations of traditional control algorithms. AstroForge plans to launch its first fully autonomous spacecraft in 2027 on Stoke Space's first rocket, backed by NASA, to gather solar data. The mission marks a significant step, as most spacecraft autonomy relies on conventional algorithms due to concerns over neural network reliability. AstroForge's co-founder, Matthew Gialich, emphasized the trade-off between building a ground network or relying on AI. The company's third spacecraft, DeepSpace-2, will launch in 2026 with Solo in 'shadow mode' for testing. Gialich stated that the Autonomy-1 mission will not include Earth communication radios, pushing for full autonomy. AstroForge's head of flight software, Armand Awad, highlighted the integration of traditional control algorithms with transformer models trained on 2,500 sensors. The AI is designed to handle tasks like anomaly resolution, such as reestablishing position tracking and addressing power issues. Despite past mission failures, the company remains committed to advancing autonomous spacecraft technology. The project represents a bold move in the space industry, aiming to reduce reliance on ground-based control systems.