My research focuses on the security, resilience, and optimization of socio-cyber-physical systems, with particular emphasis on intelligent transportation networks and urban mobility. As infrastructure systems increasingly integrate computational and sensing capabilities, they face new vulnerabilities from both cyberattacks and strategic manipulation by social agents.
Intelligent Transportation Systems
Transportation systems as socio-cyber-physical systems
I investigate the complex interplay between technology, social dynamics, and physical infrastructure through two primary research directions:
My work develops robust frameworks to maintain system integrity despite malicious attacks or hardware failures. Current projects include:
Physics-informed neural networks for nonlinear system observer design
Secure estimation and control strategies with security guarantees against cyberattacks
Experimental analysis of attack propagation in autonomous vehicle platoons
I explore how to design effective mechanisms that promote social welfare when interacting with non-myopic, strategic agents. Current investigations include:
Dynamic, budget-constrained incentive mechanisms for infrastructure sustainability
Co-learning approaches for preference estimation under information asymmetry
Network effects of eco-driving behaviors through laboratory and simulation experiments
Block diagram for monitoring and control of socio-cyber-physical systems
Resilience against faults, cyberattacks, and strategic behavior
By addressing fundamental challenges in control theory, game theory, learning theory, and dynamical systems, my research aims to enhance the resilience of critical infrastructure while ensuring security, privacy, and fairness. The ultimate goal is to develop frameworks that policymakers and system designers can implement to safeguard the reliability of the infrastructure systems that underpin modern society.
Collaborators: Munther Dahleh, Cathy Wu, Roy Dong
This project explores the design and implementation of incentive mechanisms to promote eco-driving in urban transportation networks. We develop mathematical frameworks and computational approaches that encourage drivers to adopt fuel-efficient driving behaviors while accounting for their strategic decision-making processes and the realities of information asymmetry.
The transportation sector is a major contributor to greenhouse gas emissions, accounting for approximately a quarter to a third of global emissions with urban transportation growing at an alarming rate of 2-3% annually. While long-term solutions like electric vehicles and autonomous driving technologies continue to develop, immediate action is needed to reduce emissions. Eco-driving—the practice of adopting energy-efficient driving techniques such as smoother acceleration, maintaining steady speeds, and selecting less congested routes—offers an immediate, cost-effective intervention that can reduce vehicle emissions by 10% to 45%. However, eco-driving often requires drivers to accept longer travel times, creating a misalignment between individual preferences and societal goals. This misalignment necessitates carefully designed incentive mechanisms that can persuade drivers to adopt eco-friendly behaviors while respecting their individual preferences and constraints.
The system operator provides incentives to drivers on the road to reduce their emissions by playing a Stackelberg game of incomplete information. The incentive scheme induces a Nash game among drivers, where they influence and get influenced by the driving policies of vehicles around them. (Figure by Jung-Hoon Cho)
The timeline of the eco-driving incentive mechanism.
Mathematical Framework for Incentive Design: We developed a theoretical foundation for eco-driving incentive mechanisms (EDIMs) within the context of Stackelberg games, where a transportation system operator (TSO) acts as a leader who offers incentives to drivers (followers) to encourage eco-driving behaviors.
First-Best and Second-Best Mechanisms: We characterized two approaches to incentive design:
First-Best EDIM: Optimal when drivers truthfully report their preferences, implementing recommended eco-driving levels at Nash equilibrium.
Second-Best EDIM: Robust to strategic misreporting, incorporating incentive compatibility constraints to ensure truthful behavior.
Learning-Based Approach: We created a learning-based framework that integrates reinforcement learning with regret minimization techniques to optimize incentives in complex, dynamic traffic environments where driver preferences are initially unknown.
Adjustment Mechanisms: We designed innovative contract structures with adjustment mechanisms that penalize inconsistencies in driver behavior, expanding the space of feasible truthful incentive mechanisms.
Practical Implementation: We implemented and tested our approaches in realistic traffic simulations using the SUMO platform, demonstrating significant reductions in emissions while maintaining acceptable travel times.
This research addresses a critical challenge in sustainable transportation by providing practical tools for system operators to influence driver behavior in ways that reduce environmental impact. Our findings have several important implications:
Climate Change Mitigation: By offering immediate pathways to emission reduction in urban transportation, our work contributes to broader climate change mitigation efforts without requiring major infrastructure changes.
Policy Design: Transportation authorities can use our frameworks to design more effective eco-driving incentive programs based on sound economic principles and behavioral insights.
Technological Integration: Our approaches can be integrated with existing technologies such as vehicle telematics, navigation apps, and smart infrastructure to create comprehensive eco-driving systems.
Behavioral Economics Applications: The mechanisms we developed demonstrate how economic incentives can be effectively designed to address the tension between individual preferences and collective environmental goals.
Interdisciplinary Approach: Our work showcases the power of combining tools from control theory, game theory, and machine learning to address complex societal challenges, establishing a template for future interdisciplinary research in sustainable systems.
The methodologies developed in this research extend beyond eco-driving and offer insights for incentive design in other domains where individual behaviors have significant collective impact, such as energy conservation, public health, and resource management.
M. U. B. Niazi, J.-H. Cho, M. A. Dahleh, R. Dong, and C. Wu, "Eco-driving incentive mechanisms for mitigating emissions in urban transportation," IEEE Transactions on Control of Network Systems (under review), 2024.
M. A. Dahleh, T. Horel, M. U. B. Niazi, "Mitigating information asymmetry in two-stage contracts with non-myopic agents," IFAC Workshop on Cyber-Physical Human Systems, 2024.
J.-H. Cho, M. U. B. Niazi, S. Du, T. Zhou, R. Dong, and C. Wu, "Learning-based incentive design for eco-driving guidance," Conference in Emerging Technologies in Transportation Systems (TRC-30), 2024.
M. U. B. Niazi, J.-H. Cho, M. A. Dahleh, R. Dong, and C. Wu, "Incentive design for eco-driving in urban transportation networks," European Control Conference (ECC), 2024.
Collaborators: Karl Henrik Johansson
State observers serve as critical components in modern engineering applications, including control systems, fault diagnosis, and digital twins. These mathematical tools estimate the internal state of a system using limited sensor measurements and system models—a capability essential when direct measurement of all system variables proves impractical or impossible due to sensing limitations.
This project focuses specifically on Kazantzis-Kravaris/Luenberger (KKL) observers due to their exceptional versatility with nonlinear systems and their well-established theoretical foundation. However, despite their theoretical advantages, implementing KKL observers presents significant practical challenges:
Traditional KKL observer design requires finding an injective transformation map that converts a nonlinear system into a specific required form—a mathematically complex task with no analytical solutions.
Even when this transformation is known, computing its inverse to recover estimates in the original, physically meaningful coordinates remains computationally difficult.
These limitations have restricted the practical application of KKL observers despite their theoretical potential.
We present a novel learning-based approach that addresses these fundamental challenges through:
A physics-informed neural network framework that learns both the transformation map and its inverse using synthetic data generated from system dynamics and sensor measurements.
Direct integration of the governing partial differential equation constraints into the learning process, ensuring the resulting observer satisfies the mathematical requirements.
A sequential learning methodology that first learns the transformation map and subsequently its inverse, significantly improving estimation accuracy and stability.
Our research advances the state of the art in several important ways:
Guaranteed robustness of the learned KKL observers against both learning errors and system uncertainties.
Practical applications beyond state estimation, including sensor fault detection and isolation capabilities.
Enhanced generalization performance across the entire state space, avoiding the overfitting problems common in previous approaches.
Improved training efficiency and estimation accuracy through our sequential learning architecture.
Better robustness to initialization conditions outside of training examples, addressing a key limitation of learning methods.
This research represents a significant step forward in making advanced observer technology accessible for complex nonlinear systems, with applications spanning industries from aerospace and manufacturing to energy systems and autonomous vehicles.
M. U. B. Niazi, J. Cao, M. Barreau, and K. H. Johansson, "KKL observer synthesis for nonlinear systems via physics-informed learning," Automatica (under review), 2025.
M. U. B. Niazi, J. Cao, X. Sun, A. Das, K. H. Johansson, "Learning-based design of Luenberger observers for autonomous nonlinear systems," American Control Conference (ACC), 2023.
Collaborators: Karl Henrik Johansson, Carlos Canudas-de-Wit, Alain Kibangou, Pierre-Alexandre Bliman
This is an MSCA Postdoctoral Fellowship project that aims to address critical gaps in epidemic monitoring and control systems, which were exposed during the COVID-19 pandemic. We develop an innovative framework that integrates physics-informed neural networks (PINNs) with robust system-theoretic tools to create more reliable and adaptive epidemic monitoring systems.
Traditional epidemic monitoring methods demonstrated significant limitations during COVID-19, including:
Inability to accurately predict disease evolution
Lack of robustness when handling model and data uncertainties
Difficulties accounting for complex human behavioral factors
Limited capacity to adapt to rapidly changing epidemic dynamics
These shortcomings stem from the inherent nonlinearity and heterogeneity of epidemic processes, where human behavior influences everything from disease transmission to data reporting accuracy.
The effect of lockdown on diagnosed COVID-19 cases in France during 2020.
Our Best Effort Strategy for Testing (BEST) policy uses the estimated epidemic states and parameters to determine the exact number of tests needed to mitigate an epidemic at any time. This policy can be employed at the epidemic onset. The number of tests required to control an epidemic increases exponentially as the infection spreads in the population.
Our BEST policy can be employed to minimize the burden on medical facilities during an epidemic outbreak.
We introduced a closed-loop epidemic monitoring framework with three key innovations:
Physics-Informed Neural Networks: We leverage PINNs to estimate unknown nonlinear functions in epidemic models, such as time-varying transmission rates, by combining mechanistic epidemiological models with data-driven learning.
Nonlinear Observer Design: We developed specialized nonlinear Luenberger-like observers that combine feedback and feedforward data injection, creating a self-correcting system that continuously refines predictions against real-world observations.
Robust Optimal Control: Using the reliable state estimates from our observers, we design dynamic optimal control algorithms that generate balanced policy recommendations, optimizing interventions like lockdowns, vaccination rates, and testing capacities.
This work has yielded several significant advances:
Enhanced Robustness: Our nonlinear observers effectively handle measurement and model uncertainties while providing formal guarantees of estimate accuracy.
Adaptive Modeling: The closed-loop structure enables the system to adapt to changing conditions, including social adaptations to virus spread and viral mutations.
Balanced Policy Recommendations: Our optimal control framework considers both health outcomes and socioeconomic impacts, minimizing infections and deaths while reducing economic disruptions.
Scalability Framework: We've laid the groundwork for incorporating geographic and demographic heterogeneities into large-scale networked epidemic models.
While we have made substantial progress, further development continues in:
Fully integrating geographic and demographic factors into multi-layered model architectures
Developing appropriate incentive mechanisms to improve public cooperation with health measures
Establishing partnerships with health authorities to access high-quality epidemiological data
Creating algorithms that automatically adapt these frameworks to different geographic and cultural contexts
This research significantly advances our ability to monitor and control epidemic outbreaks. By improving the accuracy and reliability of epidemic forecasting through robust feedback mechanisms, our work contributes to more effective public health decision-making and potentially more balanced interventions during future epidemics.
M. U. B. Niazi and K. H. Johansson, "Parameterization-free observer design for nonlinear systems: Application to the state estimation of networked SIR epidemics," IEEE Conference on Decision and Control (CDC), 2023.
M. U. B. Niazi and K. H. Johansson, "Observer design for the state estimation of epidemic processes," IEEE Conference on Decision and Control (CDC), 2022.
A. Alanwar, M. U. B. Niazi, and K. H. Johansson, "Data-driven set-based estimation of polynomial systems with application to SIR epidemics," 20th European Control Conference (ECC), 2022.
M. U. B. Niazi, A. Y. Kibangou, C. Canudas-de-Wit, D. Nikitin, L. Tumash, and P-A. Bliman, "Modeling and control of epidemics through testing policies," Annual Reviews in Control, 2021.
M. U. B. Niazi, A. Y. Kibangou, C. Canudas-de-Wit, and P-A. Bliman, "Optimal control of urban human mobility for epidemic mitigation," IEEE Conference on Decision and Control (CDC), 2021.
M. U. B. Niazi, A. Y. Kibangou, C. Canudas-de-Wit, D. Nikitin, L. Tumash, and P-A. Bliman, "Effective testing policies for controlling an epidemic outbreak," IEEE Conference on Decision and Control (CDC), 2021.
Collaborators: Carlos Canudas-de-Wit, Alain Kibangou, Jacquelien M. A. Scherpen
In this project, we developed a novel monitoring approach for large-scale infrastructure systems that operate as dynamical networks. This work has diverse applications, from transportation networks and smart buildings to epidemics. We employ system aggregation techniques that balance monitoring effectiveness with computational efficiency.
This work is motivated by the growing complexity of modern infrastructure systems that require practical monitoring strategies. Infrastructure system operators face challenges as systems grow in scale and complexity, creating an urgent need for monitoring approaches that can:
Function effectively within real-world computational constraints
Scale appropriately for large interconnected systems
Provide actionable insights for system operators
Illustration of the aggregation process, where our goal is to obtain a tractable representation of the large-scale network system via clustering such that each cluster is connected, information loss is minimal, and the aggregated state profiles of clusters are observable.
Aggregated monitoring of large-scale network systems with only few sensors providing local measurements.
Our research has provided several significant advances in network monitoring theory and practice:
Cluster-Based Monitoring Framework: We demonstrated that complex networks can be effectively monitored by focusing on cluster-level dynamics rather than tracking every individual node. This aggregation approach dramatically reduces computational requirements while maintaining high monitoring accuracy.
Minimum-Order Average Observers: We developed mathematical criteria for optimizing monitoring systems, identifying the minimum number of observation points and clusters needed to achieve desired accuracy levels. This contribution provides a foundation for resource-efficient monitoring across various applications.
Theoretical-Practical Integration: The research bridges sophisticated mathematical concepts with practical implementation requirements, creating monitoring solutions that are both theoretically sound and operationally viable.
The results obtained from this project have been successfully applied to the urban transportation network of Grenoble, where our methods proved valuable for:
Monitoring traffic densities across aggregated network regions
Providing optimal route recommendations based on cluster-level observations
Demonstrating how abstract mathematical principles translate into tangible transportation management solutions
The methodologies developed in this research have applications extending far beyond transportation systems. The principles of aggregated monitoring can be adapted for:
Smart city infrastructure management
Energy distribution network optimization
Environmental monitoring systems
Public health and disease surveillance networks
Our findings demonstrate that while network systems are inherently complex, they can be effectively monitored and managed through optimal aggregation and clustering approaches. This creates new possibilities for handling large-scale systems with limited resources, offering valuable tools for addressing future challenges in network management and control.
M. U. B. Niazi, X. Cheng, C. Canudas-de-Wit, and J. M. A. Scherpen, "Clustering-based average state observer design for large-scale network systems," Automatica, 2023.
M. U. B. Niazi, C. Canudas-de-Wit, and A. Y. Kibangou, "Average state estimation in large-scale clustered network systems," IEEE Transactions on Control of Network Systems, 2020.
M. U. B. Niazi, D. Deplano, C. Canudas-de-Wit, and A. Y. Kibangou, "Scale-free estimation of the average state in large-scale systems," IEEE Control Systems Letters, 2020.
M. U. B. Niazi, C. Canudas-de-Wit, and A. Y. Kibangou, "State variance estimation in large-scale network systems," 59th IEEE Conference on Decision and Control (CDC), 2020.
M. U. B. Niazi, C. Canudas-de-Wit, and A. Y. Kibangou, "Thermal monitoring of buildings by aggregated temperature estimation," IFAC World Congress, 2020.
M.U.B. Niazi, X. Cheng, C. Canudas-de-Wit, and J. M. A. Scherpen, "Structure-based clustering algorithm for model reduction of large-scale network systems," 58th IEEE Conference on Decision and Control (CDC), 2019.
M. U. B. Niazi, C. Canudas-de-Wit, and A. Y. Kibangou, "Average observability of large-scale network systems," 18th European Control Conference (ECC), 2019. (Finalist for Best Student Paper Award)
Collaborators: A. Bulent Ozgular, Aykut Yildiz
This project investigates how opinions form and evolve within complex social networks, with particular emphasis on the dynamics of belief formation across multiple interconnected issues. Our work examines the bidirectional nature of social influence, where individuals simultaneously shape and are shaped by the perspectives of those around them.
Our work is motivated by observations of real-world social systems where complete consensus rarely emerges, yet societies often achieve stable equilibria despite persistent disagreements. Traditional models have typically oversimplified these dynamics by focusing on single-issue opinion formation, overlooking the interconnectedness of belief systems, and/or neglecting the complexity of social network structures through which influence propagates.
Our analysis revealed several significant insights into opinion dynamics:
Consensus Conditions: True consensus emerges only when individuals share homogeneous conceptual frameworks about the issues at hand. Alignment in how people understand and interpret information serves as a necessary precondition for agreement.
Stable Disagreement States: When individuals operate from heterogeneous conceptual frameworks, even extensive social interaction fails to produce consensus. Instead, the system evolves toward meta-stable equilibria characterized by persistent yet non-extreme disagreement.
Polarization Mechanisms: Fundamentally opposite conceptual frameworks can trigger bifurcations in opinion space, leading to extreme polarization as an emergent property of the system dynamics.
A social network with two political parties, each with its supporters and neutral people, debating on two issues, where the supporters of party A have different conceptions of the correlation between the issues than those of party B.
When the conceptions of supporters A and supporters B about the issues are different, the Nash equilibrium in social networks tends to drive them toward more extreme, opposing viewpoints - even among individuals who were initially neutral.
These findings provide valuable insights for understanding contemporary social phenomena, from political polarization to persistent cultural divides. Our research suggests that effective strategies for social cohesion must address underlying differences in conceptual frameworks rather than merely increasing interaction frequency or communication channels.
The results challenge conventional approaches to conflict resolution by demonstrating that acknowledging fundamental differences may prove more productive than pursuing unrealistic consensus goals. This framework offers valuable perspectives for understanding and potentially addressing the complex dynamics of disagreement in today's interconnected and diverse societies.
M. U. B. Niazi and A. B. Özgüler, "A differential game model of opinion dynamics: Accord and discord as Nash equilibria," Dynamic Games and Applications, 11, pp. 137-160, 2020.
Preprint
M. U. B. Niazi, A. B. Özgüler, and A. Yildiz, "Consensus as a Nash equilibrium of a dynamic game," International Conference on Signal-Image Technology & Internet-Based Systems (SITIS), 2016.