Next-generation space missions would require self-sustaining, life-long operations of large-scale systems in highly uncertain environments with complex interactions among numerous spacecraft, robots, humans, and AIs. Hiroyasu Tsukamoto, an assistant professor at the University of Illinois at Urbana-Champaign and a research affiliate at the NASA Jet Propulsion Laboratory (JPL), is on his quest to unravel the intrinsic nature of “intelligence,” pushing the boundaries of fundamental principles in reliable, optimal, and fully autonomous decision-making in deep space exploration.
Conventionally, performance guarantees of intelligent decision-making rely heavily on strict assumptions of system-specific characteristics and uncertainties, limiting their applicability in deep space where these factors could be entirely unknown. Despite these limitations, the substantial costs involved in each space mission necessitate formal quantifications of such guarantees: only then can the mission's feasibility and sustainability be properly evaluated throughout its lifespan.
Tsukamoto has developed an innovative approach to tackle this challenge. He first proposed a data-driven approach to constructing the “Contraction Metric,” which defines a differential distance of the current systems’ performance to their ideal, exponentially decreasing in time. Thanks to its differential nature, it establishes many parallels between decision-making in linear systems and in nonlinear systems, which greatly simplify the derivation and assessment of the data-driven performance guarantees with various levels of nonlinearities.
He explored this idea further to integrate deep learning-infused intelligence into the design of the contraction metric for highly autonomous, stochastically uncertain nonlinear systems. The concept of contraction here still enables the formal quantifications of the systems’ performance guarantees in dynamically changing environments, even with radical shifts in their uncertainties. His theory thereby offers one of the most comprehensive frameworks for interpretable and trustworthy autonomy in deep space exploration, where significant uncertainty is always present in systems by nature.
Tsukamoto’s doctoral dissertation on this topic won the William F. Ballhaus Prize, the best doctoral dissertation in Space Engineering at the California Institute of Technology, in 2023. As joint projects with NASA-JPL, his theory has been successfully applied to intelligent guidance and control software for interstellar object exploration with highly unpredictable orbital properties and, more recently, to provably robust architecture for distributed multi-spacecraft networks under interactional uncertainty. He has been expanding these ideas, also at JPL, to create generalized intelligence for reliable and sustainable exploration in deep space.
Tsukamoto says, “It’s always astonishing to me that only a handful of people have landed on the moon and that even in all human history, no one has ever made it to any other planetary bodies in outer space, even though we know that they are just sitting out there waiting. It’s time to change that.” At the University of Illinois at Urbana-Champaign, home to his newly founded research group (https://hirotsukamoto.com/), he leads pioneering research and education in sustainable and autonomous space exploration, establishing the groundwork for generalized system-level autonomy. He also proactively engages in outreach activities to inspire profound societal interest in space science and engineering using his website, YouTube, and science boot camps for K12 students.
Through his endeavors in research, education, and public engagement, he envisions a world where anyone can follow their curiosity, live anywhere in space, and turn their imagination into reality to the full extent and beyond, regardless of the physical, intellectual, and temporal constraints inherent to humans.
