LPL Colloquium: Dr. Roberto Furfaro

AI for Planetary Exploration and Space Autonomy: From Near-Earth Objects to Space Domain Awareness and Beyond

When

3:45 – 4:45 p.m., Sept. 2, 2025

Where

Dr. Roberto Furfaro
Professor, Systems and Industrial Engineering
Deputy Director, S4 Space Center Director
Space Situational Awareness
University of Arizona

The exploration of planetary bodies, the defense against potentially hazardous near-Earth objects (NEOs), and the safe navigation of spacecraft in complex gravitational environments all present profound challenges at the intersection of physics, autonomy, and artificial intelligence. At the Space4 Center and Space Systems Engineering Lab, my group is advancing a unified AI framework that integrates physics-informed learning, meta-reinforcement learning, and large-scale decision support to address these challenges across domains.

We begin with our ongoing role in NASA’s NEO Surveyor mission, where we lead the development of the Target Follow-up Intelligent Decision Support System (TFO-IDSS). This system leverages Bayesian networks, physics-based orbit propagation, and LLM-powered interfaces to rank hazardous objects, assess observability, and intelligently task ground assets for recovery. By combining rigorous dynamical modeling with explainable AI, TFO-IDSS is designed to maximize the probability of detecting and tracking high-priority NEOs while providing natural language interaction for scientists and mission operators.

Building on this planetary defense effort, we extend AI methods to the broader domain of Space Domain Awareness (SDA). Physics-Informed Neural Networks (PINNs) are enabling robust orbit determination in cislunar regimes by fusing sparse telescope measurements with dynamical models, while fine-tuned Large Language Models (LLMs) provide adaptive sensor tasking, automated reasoning, and decision support in complex multi-sensor environments. Together, these tools are redefining how we acquire and exploit knowledge of the dynamic space environment.

Finally, I will discuss how meta-reinforcement learning is being developed for spacecraft Guidance, Navigation, and Control (GNC) in some of the most demanding scenarios in planetary exploration: precision lunar landing, asteroid intercept, and close-proximity operations. By training controllers that adapt online to uncertain dynamics and environmental variability, we move closer to autonomous flight systems that can learn, respond, and make decisions in real time.

Through these three interconnected thrusts, planetary defense, SDA, and autonomous GNC, we are advancing fundamental AI methods that transcend disciplinary boundaries, enabling solutions that are not only scientifically transformative but also adaptable to the continually shifting funding landscape.

View Dr. Furfaro's lecture