TILOS is the U.S. National Science Foundation (NSF) AI Institute for Learning-enabled Optimization at Scale. Now in its second phase (2026–2031), TILOS brings together faculty from University of California, San Diego, Brown University, Massachusetts Institute of Technology, National University, University of Pennsylvania, University of Texas at Austin, and Yale University. Our mission is to co-advance the next generation of AI and optimization, driven by the complexities of physically embedded real-world intelligence, while cultivating the workforce needed to shape this area. Building on Phase I breakthroughs in learning-enabled optimization, TILOS is advancing a new paradigm of optimization-driven intelligent systems spanning chips and autonomous systems.

TILOS Phase II research is organized into three thrusts:

1. Foundations: core principles for optimizing AI models to learn, reason, and generalize more effectively and efficiently, including adaptive, efficient, generalizable, trustworthy, and agentic intelligence.
2. Embodied Multi-Agent AI: how large populations of embodied agents learn, coordinate, and act under communication constraints and strict safety requirements, through world models, multi-agent optimization, and safe exploration.
3. AI Hardware Substrate: a closed-loop, AI-native hardware–software ecosystem that integrates AI for hardware design with hardware for AI, enabling next-generation platforms for critical domains such as robotics.