NSF awards $15 million to TILOS for AI research tied to the physical world

AI systems that learn, reason and coordinate in the physical world are the goal—not a result delivered by this award. The National Science Foundation has awarded $15 million over five years to the UC San Diego-led TILOS institute to pursue that broader research agenda, the university said on October 15.

The Phase II award runs from November 1, 2026, through October 31, 2031, under NSF’s National AI Research Institutes program. UC San Diego will remain the lead institution, with Yusu Wang, a professor in its Halıcıoğlu School of Data Science and Computing, serving as principal investigator.

Established in 2021, TILOS—the AI Institute for Learning-enabled Optimization at Scale—has focused on foundational work connecting learning and optimization. The renewal broadens that focus toward physically embedded intelligence: AI methods combined with optimization techniques and real-world systems. Rather than leaving the work at the level of learning methods alone, the new agenda includes how intelligent systems act together and run on computing hardware.

The institute has organized Phase II around three connected research areas. They address the methods that guide AI, the behavior of teams operating in physical environments, and the computing infrastructure needed to support them:

• Foundations: Develop AI systems that adapt, reason and generalize more effectively. The stated goals also include greater efficiency and robustness, making the underlying methods a research priority alongside their applications.
• Embodied multi-agent AI: Build systems capable of acting and coordinating as teams in the physical world. Here, “embodied” means the agents operate in physical environments, rather than coordination being considered only as a software task.
• AI hardware substrate: Design methods that help AI operate reliably and efficiently across chips and computing infrastructure. This part of the program puts the systems supporting AI inside the research agenda, alongside learning and coordinated action.

Building on the scientific foundations established during our first five years, we are expanding our focus toward intelligent systems that can learn, reason, coordinate and operate effectively in complex physical environments.

Yusu Wang, TILOS principal investigator

TILOS plans to expand pilot deployments and collaborative research in robotics, communication networks and chip design. It also plans to continue releasing open-source frameworks that connect AI and optimization. Those activities provide routes for the institute’s research to move into applications, rather than making the renewal solely a program of foundational study.

The university identifies transportation, manufacturing, healthcare, energy systems and scientific instrumentation as areas the research could serve. The intended capabilities include making real-time decisions, adapting to changing conditions and operating reliably in high-stakes settings. These are the institute’s stated application goals, not performance results from the newly funded phase.

Brown University joins TILOS as a new Phase II partner. Continuing collaborators include MIT, the University of Texas at Austin and Yale University. The renewal also adds faculty members from Brown, MIT, UT Austin and UC San Diego, widening the group contributing to the institute’s research.

Education, workforce development and outreach remain part of the program. TILOS works with industry partners to transfer research into applications and create interdisciplinary training opportunities for students. It will also continue bringing researchers, educators, industry leaders and public stakeholders together through workshops, AI Futures Roundtables and Grand Challenge events.