AI Institute in Dynamic Systems

This NSF Artificial Intelligence (AI) Research Institute in Dynamic Systems will transform research and education in fundamental artificial intelligence (AI) and machine learning (ML) theory, algorithms, and applications. The institute’s specific aim is to empower safe, guaranteed real-time learning and control of complex dynamic systems, such as autonomous vehicles, robotics, power grids, fluid flow control, digital twins, and/or advanced manufacturing.

AI/ML methods are a suite of data science algorithms that leverage diverse sensor data streams across all disciplines of science and engineering. This enables the integration of traditional modeling and computation of dynamic systems, which evolve in time and are governed by physics, with emerging AI/ML algorithmic approaches. As a result, this allows for safe, reliable, efficient, and ethical data-enabled solutions to real-time sensing, learning, prediction, and decision-making challenges.

In addition, the institute will disseminate open-source software and educational materials for the professional development of undergraduate students, graduate students, and engineering professionals alike. The institute will further develop a community-wide common task framework for evaluating a taxonomy of ML/AI algorithms on challenge data sets in physics and engineering, thus providing a broad service to the engineering AI/ML community.

The AI Institute in Dynamic Systems will transform the foundations of physics-informed AI/ML algorithms by developing the mathematical foundations in four key disciplines: (i) control theory, (ii) probability and statistics, (iii) optimization, and (iv) dynamical systems (modeling). The integration of all four of these disciplines is critical for the development of AI/ML algorithms that can be leveraged by engineered systems.

Establishing rigorous mathematical connections between these disciplines is a driving inspiration for our efforts in re-framing the foundations of AI/ML for the dynamic systems in engineering. Such foundational efforts will engender the following program thrusts: (i) the mathematical foundations of AI, (ii) grand challenge applications for AI, and (iii) a transformational educational and workforce development infrastructure for AI Engineering.

Physics-informed ML is emerging as a leading paradigm for bringing together a diverse suite of AI/ML algorithms and dynamic systems engineering, providing new capabilities in real-time sensing, learning, decision making, and predictions that are safe, reliable, efficient, ethical, and imbued with uncertainty quantification (UQ). Our team will focus on developing a general and flexible AI/ML framework to rapidly learn new physics, enforce known physical constraints, and discover them directly.

We will continuously evaluate our methods through demonstrations on real-world grand challenge applications so that the methods are physically motivated, and advances are catalyzed across multiple domains. Such an evaluation on a common task framework will engender a broad and principled taxonomy of AI/ML algorithms.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.