Abstract:
Autonomous systems can provide advantages such as access, expendability, and scaled execution in adversarial environments. However, such environments are inherently complex in the sense that they are uncertain, and data exchanges for sensing and communications can be compromised or denied. This presentation provides insights into learning for control related to uncertainty, optimality, and data intermittency that provide foundations for assured autonomous operations. Some overview of intermittent data problems will be provided along with new results for guaranteed (Lyapunov-based) deep learning methods that can be employed in real-time with no prior data. These efforts include constructive deep learning using various modern architectures.
Bio:
Professor Warren Dixon received his PhD in Electrical Engineering in 2000 from Clemson University, and after working at Oak Ridge National Laboratory as a Wigner Fellow and research staff member, he joined the University of Florida in 2004. He is a Distinguished Professor, Dean’s Leadership Professor, and Department Chair of Mechanical and Aerospace Engineering. His main research interest has been the development and application of Lyapunov-based control techniques for uncertain nonlinear systems. His work has been acknowledged by various early and mid-career awards and best paper awards. He is an ASME and IEEE Fellow for contributions to adaptive control of uncertain nonlinear systems.