Can AI make spacecraft propulsion more efficient?
Marcos Fernandez-Tous, associate professor of space propulsion and hypersonic aerodynamics at the University of North Dakota, discusses this topic.
Faculty Bio:
I currently coordinate the Space Studies Propulsion Lab within the Department of Space Sciences and offer courses on space propulsion, air-breathing engines, and hypersonic aerodynamics. My teaching emphasizes project-based learning and follows the Innovation-Based Learning methodology.
I hold a Ph.D. in Space Sciences from the University of North Dakota, where my dissertation focused on developing a novel synchronization protocol for unmanned aerial vehicles (UAVs). Additionally, I earned an M.S. in Aeronautical Engineering with a specialization in Air Traffic Management from the Polytechnic University of Madrid, Spain, completing a thesis on GPS-based approach procedures for Palma de Mallorca Airport.
My research focuses on interdisciplinary approaches to integrating artificial intelligence technologies to advance space propulsion development. Beyond academia, I actively contribute to the community of Grand Forks, where I have resided since 2018.
Transcript:
Every year, hundreds of rockets are launched into space, with numbers rising as missions to the Moon, Mars, and beyond become the new reality. But these ambitions depend on propulsion—the technology that moves spacecraft forward. Breakthroughs in this technology are essential to make interplanetary travel faster and more efficient, and artificial intelligence is emerging as a key enabler for that.
One branch of AI, reinforcement learning, teaches machines through trial and error, improving performance by interpreting outcomes and adjusting strategies. This capability is particularly valuable for complex systems that exceed human intuition, such as propulsion systems design. Reinforcement Learning can optimize flight paths to minimize fuel use and travel time, or it can assist engineers by analyzing countless variables—materials, geometry, and heat transfer—to identify configurations that maximize efficiency.
Nuclear propulsion is among the most promising concepts for deep-space missions. Fission-based systems have been tested before, while fusion remains a frontier. Nuclear thermal propulsion, which transfers heat from atomic reactions to hydrogen propellant, could dramatically reduce travel time compared to conventional propulsion. Reinforcement Learning supports this technology by optimizing reactor designs, such as those explored since NERVA program at NASA in the 60s and 70s.
Beyond design, Reinforcement Learning can manage real-time operations, including fuel consumption for adaptable missions. This flexibility is critical for spacecraft that must respond to changing priorities, such as military or multi-role satellites. The ability of Reinforcement Learning to learn and adapt makes it a powerful tool for addressing uncertainty in mission planning.
As propulsion technologies evolve, AI—and derivative tools such as reinforcement learning—will play a central role in enabling faster, safer, and more sustainable space exploration, opening pathways to destinations across and beyond our solar system.
Read More:
[The Conversation] - AI is making spacecraft propulsion more efficient – and could even lead to nuclear‑powered rockets
