Research

My research lies at the intersection of Control, AI, and Energy Systems. The overarching goal is to develop learning-enabled decision-making methods that are not only effective from data, but also interpretable, reliable, and grounded in system structure.

Topics: Data-Driven Control + Koopman Operators · Control Theory + LLMs · Machine Learning + Sustainable Energy Systems

Data-Driven Control & Koopman Operators

Modern systems are often too complex for accurate first-principles modeling, yet purely black-box models can be difficult to use for control. We study structured, data-driven representations of nonlinear dynamics, particularly Koopman-based and bilinear models, that connect learning from trajectories with predictive and optimal control.

Questions of Interest

How can we learn control-oriented representations of nonlinear systems from limited trajectory data?
How can Koopman and bilinear structure be translated into tractable predictive and optimal controllers?
What guarantees can be established for data-driven models operating in closed loop?

Selected Works

[CDC'25]	Data-enabled predictive control for nonlinear systems based on a Koopman bilinear realization. [IEEE Xplore, arXiv]
[LCSS'22]	Data-driven optimal control of bilinear systems. [IEEE Xplore, arXiv]

Control Theory & Large Language Models

Large language and multimodal models increasingly operate as agents: they receive feedback, update decisions, and interact repeatedly with users and environments. We explore how ideas from feedback control, dynamical systems, and optimization can help understand and improve these closed-loop AI systems.

Current Directions

Feedback-based methods for aligning and adapting language and vision-language models.
Control-theoretic perspectives on reliability, robustness, and safety in iterative AI decision-making.
Learning and optimization methods for agents interacting with dynamic environments.

This is an active and growing research direction in the COLAS Lab. We welcome conversations with researchers working across control, machine learning, and foundation models.

Machine Learning & Sustainable Energy Systems

The transition to low-carbon energy brings new uncertainty, decentralization, and operational complexity. We develop learning-based controllers for power and energy systems while explicitly accounting for stability, safety, communication constraints, and equitable access to network resources.

Questions of Interest

How can learned controllers retain stability and safety guarantees when deployed on physical grids?
How should distributed energy resources coordinate under limited or changing communication?
How can efficiency, resilience, and equity be incorporated into data-driven energy management?

Selected Works

[TSG'26]	Stability constrained voltage control in distribution grids with arbitrary communication infrastructures. [IEEE Xplore, arXiv]
[PSCC'24]	Unsupervised learning for equitable DER control. [ScienceDirect, arXiv]
[LCSS'23]	Learning decentralized frequency controllers for energy storage systems. [IEEE Xplore]
[LCSS'23]	Constraints on OPF surrogates for learning stable local Volt/Var controllers. [IEEE Xplore, arXiv]
[TPS'24]	Learning provably stable local Volt/Var controllers for efficient network operation. [IEEE Xplore, arXiv]
[SCL'24]	Reinforcement learning for distributed transient frequency control with stability and safety guarantees. [ScienceDirect, arXiv]

See Publications for a complete list of papers. Prospective students and collaborators are also welcome to learn more about the COLAS Lab.