RESEARCH
Past and Current Projects
6G Control via EdgeRIC: Theory and Applications
Sponsor: UCSD (Seed Grant). Duration: 2024 - 2025.
As the demand for ultra-reliable, low-latency, and high-bandwidth mobile communication systems grows, the development of 6G wireless technology presents a new frontier in network performance and capabilities. Integrating advanced control theory and reinforcement learning (RL) can address the challenges of dynamic network optimization and efficient resource management, which are critical for the deployment of 6G systems. This project seeks to develop an innovative framework that integrates Reinforcement Learning techniques in 6G wireless communication technologies, via a recently introduced monitoring and control microservice called EdgeRIC. This framework aims to optimize network efficiency, improve resource allocation, and enhance user experience in real-time, thereby pushing the boundaries of what is currently achievable in wireless communications. Our goal is to use state-of-the-art techniques in control theory to exploit multiple time scales, provide stability guarantees, and improve transient performance.
Hybrid System Models and Algorithms for Resilient Transmission Power Systems
Sponsor: Department of Energy / National Renewable Energy Laboratory. Duration: 2023 - 2025.
The increasing integration of renewable energy systems into power grids has heightened their vulnerability to instability from adversarial events such as blackouts, infrastructure failures, and external cyberattacks. Effectively modeling and mitigating these threats is essential to ensure the resilient operation of power grids. This project aims to tackle these challenges by leveraging advanced tools from switching and hybrid dynamical systems theory to develop innovative controllers and feedback mechanisms that enhance the grid's robustness, focusing on transmission systems.
Time-Certified Decision Making in Connected Autonomous Systems: Fixed-Time Equilibrium Seeking Control
Sponsor: National Science Foundation. Duration: 2023 - 2026.
Traditional smooth feedback algorithms and controllers, including Lipschitz-based methods with adaptive and real-time learning capabilities, often experience prohibitively slow convergence rates. This project aims to enhance the transient performance of adaptive systems by integrating fixed-time (FxT) stability guarantees through non-smooth algorithms. FxT properties ensure convergence to a desired point or set of interest within a pre-defined time, regardless of the algorithm’s initialization. The project seeks to advance these tools in challenging dynamic decision-making problems (e.g., with time-varying and dynamic plants in the loop) where conventional approaches fall short. By developing FxT tools for variational inequalities, singular perturbations, averaging theory, and interconnected systems, the project will open the door to new feedback architectures and algorithmic designs with desirable FxT properties.
Stability Preserving Adaptive Load Shedding with Energy Justice Aware Actuators
Sponsor: Department of Energy / Sandia National Laboratory. Duration: 2023 - 2026
The increasing controllability of the power grid through smart sensing and actuation devices presents an opportunity to integrate socioeconomic metrics that can help mitigate unintended negative societal outcomes. A common example is load-shedding during generation faults or sudden spikes in power demand, which can lead to the (repetitive) disconnection of power in regions with vulnerable communities. How can we design smarter, more resilient algorithms that consider societal factors to prevent such negative impacts while maintaining the grid's stable and efficient operation? This project addresses this challenge by employing distributed optimization techniques and "criticality functions" to solve the distributed load-shedding problem across discrete sets of loads and mitigating negative effects in vulnerable communities.
Nonsmooth Control Systems for Societal Networks with Data-Assisted Feedback Loops: Theory and Algorithms
Sponsor: National Science Foundation. Duration: 2022 - 2027
A major shift --perhaps equal parts lamentable and exciting-- has taken place over the last decade, and likely not to be reversed during our lifetimes, from model-based (physics-based) to model-free (data-based) approaches to feedback control and optimization. This shift is necessary for the adoption of automated decision-making at scale in engineering, socio-economic, biological, and other complex networked systems. The success of machine learning (ML) --the leading exponent among data-based tools-- in many applications has emboldened its consideration for systems with feedback loops and in the presence of dynamics. This, in turn, has led to the awareness of the pitfalls of model-free decision-making without stability and robustness guarantees. Motivated by these challenges. this project seeks to study the systematic incorporation of data-enabled and ML-based feedback mechanisms into nonlinear and hybrid controllers, where traditional assumptions of data-driven control may not hold. Special attention is given to the dynamic interactions that can emerge from the data associated with each subsystem and their impact on the overall system's stability properties.
Clusters of Flexible PV-Wind-Storage Hybrid Generation
Sponsor: Department of Energy / Sandia National Laboratory. Duration: 2022 - 2023.
Virtual Power Plants (VPPs) have emerged as a modern real-time energy management architecture that seeks to synergistically coordinate an aggregation of renewable and non-renewable generation systems to overcome some of the fundamental limitations of traditional power grids dominated by synchronous machines. The virtual (i.e., software-based) component of a VPP, combined with the power plant (i.e., physics-based) components of the power grid, make VPPs prominent examples of cyber-physical systems, where both continuous-time and discrete-time dynamics play critical roles in the stability and transient properties of the system. This project seeks to improve the performance of VPPs, with a focus on frequency control. To improve transient performance (reduced overshoot and improved convergence time) in frequency control we introduce Droop Reset Integral Control, which incorporates virtual resets to dissipate energy in the control system. The controllers have theoretical stability guarantees (via hybrid control tools) and are validated in a high-fidelity simulation model developed by Sandia National Laboratories.
Multi-Time Scale Stochastic Hybrid Control: Coordinated Set-Seeking on Manifolds
Sponsor: Air Force Office of Scientific Research. Duration: 2022 - 2025.
This project aims to develop a novel set of methodologies for the analysis and design of multi-time scale controllers in deterministic and stochastic hybrid dynamical systems evolving on smooth manifolds. These systems are of significant interest to the Air Force due to their relevance in applications such as unmanned aerial vehicles, multi-agent robotic systems, mobile networks, human-machine interactions, and decision-making algorithms, among others. The proposed research aims to develop new analytical tools and novel architectures of control algorithms that systematically exploit multiple time scales in the closed-loop dynamics of the system. The efforts are focused on two particular domains: singular perturbations and (classic and high-order) averaging theory. Applications of interest include seeking algorithms for model-free optimization and regulation (via hybrid vibrational control) of complex dynamical systems evolving under geometric constraints.
High-Performance Adaptive Hybrid Feedback Algorithms for Real-Time Optimization and Learning in Networked Transportation Systems
Sponsor: National Science Foundation. Duration: 2020 - 2022.
The recent data revolution is driving transformative changes across various engineering systems that depend on intelligent, real-time feedback for proper operation. Network transportation systems (NTS) are among the areas poised to benefit significantly from these advances by integrating data-driven feedback control and advanced optimization techniques to safely reduce road congestion while enhancing network efficiency via innovative control algorithms. This project aims to advance these objectives by developing feedback optimization algorithms for nonlinear time-varying systems. The proposed algorithms can integrate recorded data into the closed-loop system to minimize online exploration. Particular focus is given to adaptive dynamic pricing, learning-based incentives, congestion control, and adaptice traffic light systems.
Deception and Counter-Deception: A Game Theoretical Framework for Control
Sponsor: CU Boulder (Seed Grant). Duration: 2021 - 2022.
Game theory offers a powerful framework for analyzing interconnected multi-agent engineering systems involving both cooperative and uncooperative entities. Traditional game-theoretic equilibrium concepts, such as Nash equilibria, are based on the assumption that all players follow certain action rules and share truthful information about their states and beliefs. However, these assumptions may break down when action rules involve real-time adaptive and learning-based mechanisms, leading agents to potentially learn 'false' beliefs. Such beliefs can, in turn, be systematically injected and shaped by agents with privileged information. This project aims to model these interactions through "deception theory", characterizing the outcomes that arise from different game-theoretic common learning rules when certain players possess privileged information that allows them to manipulate the beliefs of others.
Real-Time Optimization of Energy Systems with Unknown Models: Accelerated Concurrent Learning
Collaborator: Robert Bosch, LLC. Duration: 2020 - 2022
Concurrent learning (CL) is a widely used technique in adaptive systems, particularly when persistence of excitation (PE) conditions are difficult to meet, but historical data is available for learning purposes. However, traditional CL techniques can experience slow convergence when the information contained in the recorded data is "weak". How can we enhance the transient performance of CL techniques using cutting-edge control methods? To address this, we propose novel CL algorithms that incorporate momentum (HACL), finite-time convergence (FTCL), fixed-time convergence (FxTCL), and Hamiltonian-based algorithms with adaptive restarting (HHCL). For all these methods, we provide stability and robustness guarantees and compare their performance to traditional CL techniques in the context of parameter estimation problems for batteries.
Cooperative Learning-based Control for Multi-Vehicle Systems
Collaborator: Mitsubishi Electric Research Laboratories. Duration: 2019 - 2021
In multi-vehicle autonomous systems operating in unknown or adversarial environments, achieving both target regulation (i.e., stabilization) and obstacle avoidance simultaneously presents a significant challenge. Even in single-vehicle systems, smooth time-invariant feedback controllers based on navigation or barrier functions have been shown to be highly vulnerable to even minor jamming signals, which can destabilize the closed-loop system, stabilize spurious equilibria in the operational space, or induce deadlocks. When the location of the target is unknown, adaptive navigation dynamics may face similar limitations. This project studies controllers that can overcome these challenges by incorporating adaptive and learning mechanisms in hybrid controllers for obstacle avoidance and source-seeking missions. The proposed controllers can use real-time and past-recorded data, as well as "cooperative exploration strategies" to facilitate the discovery of the source. Safety considerations are also enforced via projections in the control dynamics.
Coopetitive Mobile Energy Network Systems: Achieving Robust Decentralized Autonomy via Data-Driven Intelligent Algorithms and Interacting Architectures
Sponsor: CU Boulder (Seed Grant). Duration: 2019 - 2020.
Resilient coordination in networked multi-agent systems is crucial for implementing control and optimization algorithms with guaranteed stability. In this project, we investigate coordination and synchronization challenges in networks of pulse-coupled oscillators across various graph topologies and connectivity conditions. Our focus is on scalable algorithms that can accommodate stochastic communication protocols, as well as scalable "richness" conditions for data distributed among multiple agents, which is used for model learning via parameter estimation algorithms with spatial averaging (i.e. consensus-like methods).
