Completed Projects

"NSF CAREER: Distributionally Robust Learning, Control, and Benefits Analysis of Information Sharing for Connected Autonomous Vehicles", funded by NSF.

"SCC-IRG Track 1: Socially Informed Services Conflict Governance through Specification, Detection, Resolution and Prevention", funded by NSF.

"CPS: Small: COLLAB: Improving Efficiency of Electric Vehicle Fleets: A Data-Driven Control Framework for Heterogeneous Mobile CPS", funded by NSF.

"S&AS: FND: COLLAB: Adaptive and Cognizant Vehicular Sensing and Control for Fleet-Oriented Systems in Smart Cities", funded by NSF.

"Robust Control Protocol Synthesis and Safe Learning for Connected Autonomous Vehicles", May 2019 - June 2020, funded by UConn Research Excellence Program (REP).

"S&CC: Empowering Smart and Connected Communities through Programmable Community Microgrids", funded by NSF.

"Modeling, Analysis and Anomaly Detection for Cyber Secure Eversource Power Distribution Networks", funded by Eversource Energy, a northeast energy company.

"Energy Management Systems for Subtractive and Additive Precision Manufacturing", funded by Clean Energy Smart Manufacturing Innovation Institute (CESMII).

Towards Trustworthy Embodied AI: Uncertainty Quantification, Safe Robot Learning, and Foundation Models

Embodied AI promises to endow robots with intelligence in the open physical world, yet safety, robustness, and trustworthiness remain the key barriers to deploying learning-based systems in dynamic, unstructured, and safety-critical environments. This line of work focuses on a unified research agenda toward safe and trustworthy embodied AI, built on three pillars. First, we develop uncertainty quantification methods for 2D/3D perception and trajectory prediction, delivering calibrated uncertainty estimates that enable risk-aware planning and control. Second, we advance safe and robust multi-agent reinforcement learning, integrating control theory-based safety shields for deep reinforcement learning, and validate zero-shot, collision-free coordination of the trained policies on physical robot platforms. Third, in the era of foundation models, we propose efficient fine-tuning via dynamic routing of LoRA experts, scalable low-rank RL training, hallucination mitigation for multimodal LLMs, and semantically enhanced observations for robust vision-language-action manipulation. Together, these works chart a path toward embodied systems that perceive honestly, decide safely, and act reliably in the complex and ever-changing physical world.

"NSF CAREER: Distributionally Robust Learning, Control, and Benefits Analysis of Information Sharing for Connected Autonomous Vehicles", funded by NSF.

"Robust Control Protocol Synthesis and Safe Learning for Connected Autonomous Vehicles", May 2019 - June 2020, funded by UConn Research Excellence Program (REP).

Relevant publications: [α-OCC_TMLR'26] [Hallucination_COLM'26] [LD-MoLE_ICLR'26] [YOLO-MARL_IROS'25] [STL-MARL_IROS'25] [SafeMARL_ICRA'25] [CUQDS_AAAI'25] [RobustMDP_NeurIPS'24] [ConstrainedRL_ICML'24] [FedRL_ICML'24] [CoMOT_RA-L'24] [StateAdvMARL_TMLR'24] [RobustMARL_TMLR'23] [MARLPlanning_TITS'23] [SafeMARL_ICRA'23] [UQPerception_ICRA'23] [FLTP_IROS'23] [ICRA'22] [SafeMARL_Journal22] [CDC'19]

Preprints: [Sim2RealMARL_arXiv'25] [AdvUQDetection_arXiv'25] [RobustConstrainedRL_arXiv]

MARL and Robust Optimization for Sustainable Mobile CPS

Detail of the project and related publications can be found through the project webpage. Ubiquitous sensing in smart cities enables large-scale multi-source data collected in real time, and calls for a paradigm shift toward data-driven cyber-physical systems (CPS) that integrate optimization, control, and machine learning. This research agenda centers on real-time robust resource allocation for urban-scale vehicular systems, such as taxis, ride-sharing, and electric vehicle fleets. On one front, we design data-driven dynamic robust optimization frameworks that match vehicle supply to both current and predicted future demand, providing probabilistic guarantees for the system's worst-case and expected performance, complemented by dynamic pricing and hierarchical carpool algorithms for travel-time reliability during peak hours. On another front, we develop fleet-oriented sensing and control frameworks that seamlessly integrate historical and real-time data within a fleet, exploiting spatiotemporally-correlated contextual information among vehicles (e.g., vehicular mobility, service demand, disruptive events) to enable reconfigurable fleet-wide coordinated sensing, mobility-phenomenon modeling, and robust dispatching. For electric vehicles such as E-buses and E-taxis, we co-design charging and passenger pick-up algorithms. Based on real-world taxi operational data, these methods have been shown to shorten passenger waiting time, reduce cruising mileage, and improve operator revenue.

"CPS: Small: COLLAB: Improving Efficiency of Electric Vehicle Fleets: A Data-Driven Control Framework for Heterogeneous Mobile CPS", funded by NSF.

"S&AS: FND: COLLAB: Adaptive and Cognizant Vehicular Sensing and Control for Fleet-Oriented Systems in Smart Cities", funded by NSF.

Relevant publications: [IEEE_TITS23] [EVBalancing_IROS'23] [EVRebalancing_IROS'23] [TCPS20] [IROS'20] [CDC'19] [ICCPS'18] [CDC'17] [TCST19] [ICCPS'17] [TASE16] [CDC'15] [ICCPS'15]

Attack Detection and Resilient Control of Cyber-Physical Systems

Cyber-physical systems form a ubiquitous, networked, computing substrate that underlies much of modern technological society, such as autonomous vehicles and intelligent transportation systems, power networks and smart grids, and smart manufacturing systems. Researchers have shown that these kinds of networked embedded systems are vulnerable to remote security attacks, and such attacks can cause physical damage while hiding their effects from system state monitors or controllers. The requirement of designing secure networked control systems introduces new challenges: it is necessary to design attack-resilient control schemes and architectures capable of dealing with attacks on the environment of the controller, including attacks on sensors, actuators, and communication media. For a general linear system model with networked sensors and actuators, we design a hybrid-state, finite-horizon, zero-sum stochastic game approach to obtain an optimal control policy that balances the security overhead with control cost; the controller switches between subsystems in the presence of different types of attacks. We also design time-varying coding techniques with respect to sensor outputs for detecting stealthy data injection attacks on communication channels. For power systems, we analyze the effects of power botnet attacks on distribution networks, and design learning-based attack detection and localization algorithms.

"Modeling, Analysis and Anomaly Detection for Cyber Secure Eversource Power Distribution Networks", funded by Eversource Energy, a northeast energy company.

"Energy Management Systems for Subtractive and Additive Precision Manufacturing", funded by Clean Energy Smart Manufacturing Innovation Institute (CESMII).

Relevant publications: [GridDefense_TSG'23] [GridDefense_ISGT'23] [Automatica18] [TCNS16] [CDC'14] [CDC'14] [CDC'13]

Quantum Cryptography and Quantum Computing

Our quantum research spans quantum cryptography and quantum computing. On the cryptography side, we analyze quantum key distribution (QKD) protocols through a game-theoretic framework. We propose a general-purpose framework that, to our knowledge, is the first to enable critical security computations, demonstrated on several protocols and attack scenarios. The method requires fewer assumptions on the honest users and applies to practical, real-world devices. In particular, we show that if an adversary is rational as opposed to simply malicious, users may increase QKD key generation rates beyond the standard adversarial model. On the computing side, we develop variational quantum algorithms that solve constrained optimization problems on hybrid qubit-qumode quantum devices. By encoding Quadratic Unconstrained Binary Optimization (QUBO) instances across multiple qumodes weakly coupled to a single qubit, and applying the Echoed Conditional Displacement Variational Quantum Eigensolver (ECD-VQE), our approach achieves higher-quality solutions with dramatically fewer resources than qubit-only architectures, demonstrating the potential of hybrid quantum platforms for NP-hard optimization and quantum chemistry problems.

Relevant publications: [JCTC'26] [Quantum Inf. Process'20] [GameSec'18]

Wireless Control Networks and Wireless Sensor Networks

Wireless Control Network (WCN) is a network architecture in which the network itself acts as a distributed, structured dynamical compensator. Since radio communication quality between low-power sensor devices varies with time and environment, static transmission power control is often ineffective in practice. We design ATPC, a lightweight, feedback-based adaptive transmission power control algorithm in which each node models the correlation between transmission power and link quality for each neighbor. Real-world experiments show greater energy savings and robustness even as the environment changes over time.