AI in Autonomous Space Systems: Combining Strengths with Physics-Based Guidance
What is the right role for AI in autonomous space systems? My view is that the most promising path is not to replace physics-based guidance, navigation and control with AI, but to combine their strengths. At the upcoming 3rd IAA Conference on AI for Space, I’ll present a keynote on our work on AI for Distributed Space Systems and a principle that has increasingly shaped our research:
- AI proposes, accelerates and generalizes.
- Physics-based estimation and optimization constrain, refine and provide safety.
I’ll show how this “golden rule” plays out across several generations of our work: from neural pose estimation and reconstruction of unknown spacecraft, to transformer-accelerated trajectory optimization, language-grounded intent-aligned safe guidance, and our latest steps toward scaffolded foundation models and agentic autonomy for space.
The broader goal is scalability: moving beyond algorithms designed for one narrowly defined scenario toward spacecraft that can perceive unfamiliar objects, interpret higher-level intent, reason over physically meaningful actions, and coordinate as distributed systems, but without giving up the structure and guarantees that spaceflight demands. If you are working at the intersection of AI, autonomy, GNC, robotics, or distributed space systems, I hope you’ll join us in Jeju (South Korea). I’m looking forward to the technical discussions and especially to hearing where others see the right boundary between learned intelligence and model-based engineering.
3rd IAA Conference on AI for Space https://iaa-aispace.org/