Project Grant 2231350

This three-year, $240,000 Project Grant from the National Science Foundation's Division of Electrical, Communications and Cyber Systems, under the Engineering program (CFDA 47.041), will support collaborative research between Columbia University and the University of Pennsylvania on scalable and communication-efficient learning-based distributed control. The researchers will develop a foundational and integrated theory of distributed learning-enabled control and approximated distributed optimization. This integrated approach aims to design controllers that can be realistically deployed at societal scales in intelligent infrastructure systems, addressing current barriers to practical use of safety-constrained real-time distributed control related to speed, nonlinearity, and lack of structured models. The work will cover system identification incorporating federated and statistical learning; speeding up distributed predictive control using distributed imitation and federated learning tools; and designing distributed controllers robust to uncertainty from learning, numerics, and communication failures. Experimental validation will occur on robotic platforms at Penn and plasma fusion tokamak data from Columbia. Educational outcomes will include integrating the research into graduate courses at both institutions.