AM Seminar: Model-based Data-driven Computational Optimal Control

Speaker Name: 
Qi Gong
Speaker Title: 
Professor
Speaker Organization: 
University of California, Santa Cruz
Start Time: 
Monday, January 13, 2020 - 4:00pm
End Time: 
Monday, January 13, 2020 - 5:00pm
Location: 
BE 372
Organizer: 
Abhishek Halder

Abstract

This talk presents some recent developments in computational optimal control using model-based data-driven approaches for both closed-loop and open-loop solutions. For optimal feedback (closed-loop) control, a machine learning method for solving Hamilton-Jacobi-Bellman (HJB) equations will be discussed. In this method, we model solutions to HJB equations with neural networks (NNs) trained on data generated in a causality-free fashion. Since the method does not require any state space discretization over a grid, it offers a great potential to alleviate the curse of dimensionality. Training of the neural network is made more effectively and efficiently by leveraging the known physics of the problem and using the partially trained NN to aid in adaptive data generation. In the second part of the talk, we present a generalized optimal control formulation with explicit consideration of uncertainty. Such problem formulation can be applied to a variety of control applications including ensemble control, optimal search, and motion planning of autonomous vehicles. We will present a flexible numerical framework for producing solutions using a variety of discretization schemes, which can be catered to numerical needs; and analytic conditions for analysis in the form of a Pontryagin-like minimum principle. Applications on engineering problems including optimal search of unknown targets, and stochastic motion planning of autonomous vehicles will be presented.

Bio

Qi Gong is a Professor and Department Chair in Applied Mathematics at Baskin School of Engineering, University of California, Santa Cruz. He received his B.S. degree in Automation from Dalian University of Science and Technology in 1996, his M.S. in Automatic Control from Southeast University in 1999, and his Ph.D. in Systems and Control Engineering from Case Western Reserve University in 2004. His research centers on computational methods for nonlinear optimal control, trajectory optimization, optimal control theory, and engineering control applications.

Event Type: 
Event