����ֱ��

Data is everywhere in today’s world.

Assistant Professor Jorge Poveda has received a National Science Foundation that will allow him to explore how all of that data can be used to control systems — from robotic networks to the power grid.

Poveda explained that data-assisted control has become a relevant topic as machine learning has become more and more ubiquitous.

“The question is, can we systematically incorporate this data into our controllers to perhaps improve the performance of the system or be able to achieve something that is unachievable without data,” he said. “If I'm using elements from machine learning that incorporate real-time and recorded data, how can I incorporate these elements into my controllers to overcome fundamental limitations of model-based approaches?”

Specifically, Poveda plans to look at data-assisted control in non-smooth dynamical systems — an area that requires more research to create robust, stable systems with high-performance guarantees.

“The question is not easy because most of the results in the data-driven control literature rely on smoothness in some sense,” Poveda said. “They exploit smoothness in the controller to obtain robustness guarantees and stability guarantees.”

While he intends to start by developing models for data-assisted switching controllers, which switch between sub-algorithms to achieve better performance, he also plans to extend those results to networks and multi-agent systems.

“In large-scale networks, we cannot ask that every node of the network has access to unlimited data, but that data is going to be distributed between the different nodes,” he said. “The question becomes how should these nodes talk to each other? How should they share their data in real time while preserving closed-loop stability of the underlying dynamical systems?”

Poveda is also interested in incorporating insights from game theory. He explained that in many societal engineering systems, decision-makers can have conflicting interests. He wants to explore whether it’s possible for one agent to dynamically manipulate its data at the expense of others.

“There are many important questions there because we envision that in the future, many engineering systems will follow these types of rules,” he said.

He used the example of transportation systems and energy markets. Experts envision that during the next decade, people are going to be able to buy and sell energy to the power grid based on their transportation needs.

“So now we have these competitive markets with potentially adversarial behaviors such as collusions between some buyers and sellers,” he said. “If we have companies or users implementing data-driven or learning-based algorithms to update their actions, how robust will these algorithms be to adversarial data that is dynamically manipulated by deceptive entities with ‘inside’ information?”

Poveda and his graduate students are also eager to share their control systems knowledge with the next generation of diverse STEM students. As part of the CAREER proposal, they are planning several activities to recruit, mentor and retain students from underrepresented backgrounds, including K-12 summer camps, Discovery Learning Apprenticeships and an annual regional workshop on control.

The lab members will start by participating in an NSF science communication workshop to learn more about making the complicated subject accessible to those outside the field, Poveda said.

“I think it's exciting that the students in my group are going to be involved,” he said. “Those are skills they can use in the future, and mentorship is something that we value a lot.”

CAREER awards provide approximately $500,000 over five years for junior faculty members “who have the potential to serve as academic role models in research and education and to lead advances in the mission of their department or organization.”��Six faculty members from the College of Engineering and Applied Science received NSF CAREER Awards��in 2022.