Your browser is unsupported

We recommend using the latest version of IE11, Edge, Chrome, Firefox or Safari.

Apr 9 2021

Learning For Estimation and Control Using Digital Twins


April 9, 2021

11:00 AM - 12:00 PM


online: pwd=UW5GK0lEb0FmZWYyeHlFOHlrekdCQT09


Chicago, IL 60607

Learning For Estimation and Control Using Digital Twins

Presenter: Ankush Chakrabarty, Mitsubishi Electric Research Labs

Abstract: Advancements in predictive modeling informed by physics and data, along with acceleration of numerical computing, has resulted in the development of high-fidelity software replicas of modern engineering systems. These replicas, referred to as digital twins, often comprise complex software modules that are not amenable for control system design using classical techniques. In such circumstances, learning from simulation data generated by digital twins opens a path towards controller and estimator design despite the lack of control-oriented (e.g., state-space, transfer function) models. This talk posits that leveraging simulation data obtained from digital twins enables learning based control and estimation architectures that can provide solutions to critical real-world problems arising in a variety of engineering domains. In particular, we present recent work at MERL on (1) scalable Bayesian optimization for optimizing the performance of industrial-grade building control systems; and, (2) control under safety constraints using neural-approximated reachable sets for safety-critical applications such as autonomous vehicles.

Speaker bio: Ankush Chakrabarty received his PhD in electrical and computer engineering from Purdue University as a Ross Fellow in 2016. Subsequently, he was a postdoctoral fellow at Harvard University, where he developed control algorithms for a wearable artificial pancreas.

Chakrabarty now works as a research scientist at Mitsubishi Electric Research Labs, where he works on machine learning and control engineering for applications including buildings, vehicles, and factory automation. He has published over 50 peer-reviewed articles and filed over 15 patents, contributing to areas including machine learning for control, approximate nonlinear model predictive control, unknown input estimation, linear matrix inequalities, and embedded (wearable/implantable) medical devices. He is a senior member of the IEEE. He has an Erdös number of 4 and competitively writes short science fiction (a practice he avoids extrapolating to technical papers).

Faculty Host: Amit Ranjan Trivedi,



Department of Electrical and Computer Engineering

Date posted

Apr 6, 2021

Date updated

Apr 6, 2021