AstroForge Bets on AI to Solve Deep Space Communication Limits

Traditional space missions rely on rigid, rule-based algorithms due to the inherent unpredictability of neural networks. AstroForge, a startup backed by $56 million, aims to break this reliance after struggling to maintain contact with its Odin spacecraft in 2025. Facing the prohibitive $200 million cost of building a global ground station network, CEO Matthew Gialich opted to pursue onboard intelligence that can diagnose and resolve anomalies independently.

The Solo system integrates traditional control algorithms with a transformer model trained on data from 2,500 onboard sensors. Rather than chasing general artificial intelligence, the team is building a constrained system designed for specific tasks, such as recognizing power anomalies or recalibrating navigation sensors. According to head of flight software Armand Awad, the model acts as an intelligence layer capable of making corrective decisions in real time.

Before the company commits to a fully autonomous flight, it will test the software in shadow mode on the DeepSpace-2 mission, scheduled for late 2026. This trial serves as a precursor to the 2027 Autonomy-1 mission. While Gialich currently plans to launch that vehicle without a radio receiver, he admits the final decision remains fluid. The goal is to prove that a spacecraft can operate as a self-correcting agent, turning a technical necessity into a new paradigm for asteroid mining and deep space exploration.