Adversarial Pursuits in Cislunar Space

Paper: [AIAA SciTech 2026]

Collaborators: Quentin Rommel, Gregory Falco, Ufuk Topcu

We study pursuit–evasion between two spacecraft in cislunar orbit, formulated as a zero-sum dynamic game in the circular restricted three-body problem. The evader avoids capture by adjusting its orbital phasing and exploiting the orbit's stable and unstable manifolds.

Coordinated UAV Beamforming and Control for Directional Jamming and Nulling

Paper: [arXiv 2025]

Collaborators: Brian M. Sadler, Ufuk Topcu

We study the coordinated beamforming and control problem for a UAV equipped with two omnidirectional antennas, which it uses to jam an eavesdropper while leaving a friendly client unaffected. Our formulation jointly optimizes position, antenna orientation, and phase to realize directional jamming, such as selective GPS denial, while minimizing interference to protected receivers.

Optimal Satellite Maneuvers for Spaceborne Jamming Attacks

Paper: [AIAA SciTech 2026]

Collaborators: Quentin Rommel, Brian M. Sadler, Ufuk Topcu

We consider a rogue satellite in low Earth orbit that jams the uplink communication between another satellite and a ground station. To achieve maximal interference with minimal fuel consumption, the jammer carefully maneuvers itself relative to the target satellite's antenna. We cast this maneuvering objective as a two-stage optimal control problem, involving i) repositioning to an efficient jamming position before uplink communication commences; and ii) maintaining an efficient jamming position after communication has started.

Noncooperative Equilibrium Selection via a Trading-based Auction

Paper: [arXiv 2025]

Collaborators: Jaehan Im, Daniel Delahaye, Ufuk Topcu, David Fridovich-Keil

We propose an algorithm named trading auction for consensus (TACo), a decentralized approach that enables noncooperative agents to reach consensus without communicating directly or disclosing private valuations. TACo facilitates coordination through a structured trading-based auction, where agents iteratively select choices of interest and provably reach an agreement within an a priori bounded number of steps.

Approximate Feedback Nash Equilibria with Sparse Inter-Agent Dependencies

Paper: [arXiv 2025]

Collaborators: Xinjie Liu, Jingqi Li, Mustafa O. Karabag, Jesse Milzman, David Fridovich-Keil, Ufuk Topcu

We propose a regularized dynamic programming approach for finding sparse feedback policies that selectively depend on the states of a subset of agents in dynamic games. The proposed approach solves convex adaptive group Lasso problems to compute sparse policies approximating Nash equilibrium solutions. We prove the regularized solutions' asymptotic convergence to a neighborhood of Nash equilibrium policies in linear-quadratic (LQ) games. Further, we extend the proposed approach to general non-LQ games via an iterative algorithm.

Deception Against Data-Driven Linear-Quadratic Control

Paper: [arXiv 2025]

Collaborators: Aris Kanellopoulos, Kyriakos G. Vamvoudakis, Ufuk Topcu

We consider a setting where an adversary tries to learn the optimal linear-quadratic attack against a system, the dynamics of which it does not know. On the other end, a defender who knows its dynamics exploits its information advantage and injects a deceptive input into the system to mislead the adversary. We design the defender's deceptive input stratigically, forcing the adversary to learn, as closely as possible, a pre-selected attack that is different from the optimal one.

Adversarial Observability and Performance Tradeoffs in Optimal Control

Paper: [arXiv 2025]

Collaborators: Ufuk Topcu

We develop a feedback controller that minimizes the observability of adversarial sensors in a linear system while satisfying strict closed-loop performance constraints. We measure sensor effectiveness using the trace of the observability Gramian and its inverse, capturing both average and least observable state directions. We establish theoretical lower bounds on these metrics under performance constraints and optimize them using semidefinite programming.

Adversary's observer error using the nominal optimal controller.

Adversary's observer error using the proposed low-observability controller.

Input-Output Data-Driven Sensor Selection for Cyber-Physical Systems

Paper: [Automatica 2025]

Collaborators: Kyriakos G. Vamvoudakis

We propose the first model-free sensor selection framework for unknown cyber-physical systems (CPS). Given a large pool of available sensors, our goal is to select a subset that maximizes the system's observability. Unlike existing methods that rely on explicit system models, our approach performs this optimization directly from input–output data, without prior knowledge of the system dynamics.

A Physics-Informed Learning Framework to Solve the Infinite-Horizon Optimal Control Problem

Paper: [Int. J. Robust & Nonlinear Control 2025]

Collaborators: Kyriakos G. Vamvoudakis

We propose a physics-informed neural network (PINN) framework to solve the infinite-horizon optimal control problem for nonlinear systems. To address the challenge of multiple solutions in the corresponding steady-state Hamilton–Jacobi–Bellman (HJB) equation, we instead apply PINNs to a finite-horizon variant of the HJB equation that admits a unique solution, which uniformly approximates the optimal value function as the horizon length increases.

On the Effect of Clock Offsets and Quantization on Learning-Based Adversarial Games

Paper: [Automatica 2024]

Collaborators: Aris Kanellopoulos, Kyriakos G. Vamvoudakis, Jerome Hugues

We consider an off-policy iterative reinforcement learning (RL) algorithm for continuous-time systems, which uses input–state data to approximate the Nash equilibrium of a zero-sum game. However, the data employed by this algorithm are temporally inconsistent, as each sample originates from a slightly different past time instant. We prove that, under a Lipschitz continuity assumption and provided that these timing inconsistencies remain below a prescribed threshold, the off-policy iterative RL algorithm still converges epsilon-closely to the desired Nash policy.

Input-Constrained Prescribed Performance Control for High-Order MIMO Uncertain Nonlinear Systems

Paper: [IEEE TAC 2024]

Collaborators: George A. Rovithakis

We develop an approximation-free prescribed-performance controller for uncertain nonlinear systems that remains stable under strict actuator saturation. The controller dynamically relaxes and restores the reference to prevent internal instability and ensure bounded closed-loop signals. We also quantify the resulting performance distortion, providing explicit guarantees on tracking accuracy under input constraints.

Data-Driven Actuator Allocation for Actuator-Redundant Systems

Paper: [IEEE TAC 2024]

Collaborators: Kyriakos G. Vamvoudakis, Zhong-Ping Jiang

We consider the problem of optimally augmenting an actuator redundant system with additional actuators, so that the energy required to meet a given control objective is minimized. We study two cases: one where the control objective is unknown, and another where it is defined by a linear state-feedback law. In the latter case, we relax the need for explicit system models by using trajectory data to iteratively approximate the antistabilizing solution of an associated algebraic Riccati equation.

Learning-Based Actuator Selection for Increased Attack Resilience of Uncertain Systems

Paper: [Automatica 2024]

Collaborators: Kyriakos G. Vamvoudakis

We consider a security-aware and learning-based actuator selection formulation for cyber-physical systems. A set of actuators is chosen to be used by the system so that a metric of controllability and actuation attack resilience is maximized, given only partial knowledge of the physics of the system.

Recursive Reasoning with Reduced Complexity and Intermittency for Non-Equilibrium Learning in Stochastic Games

Paper: [IEEE TNNLS 2023]

Collaborators: Kyriakos G. Vamvoudakis

We develop a computationally and communicationally efficient framework for decision-making in nonequilibrium stochastic games. By modeling agents with bounded rationality through level-k and cognitive-hierarchy reasoning, we capture more realistic behaviors than standard Nash equilibria. We introduce recursive and parallel algorithms that compute these boundedly rational strategies efficiently.

Concurrent Receding Horizon Control and Estimation Against Stealthy Attacks

Paper: [IEEE TAC 2023]

Collaborators: Kyriakos G. Vamvoudakis

We present a game-theoretic framework for cyber-physical systems in which a defender mitigates attacks from an intelligent adversary that exploits system uncertainty to remain undetected. The defender jointly performs receding-horizon control and estimation, optimizing a robust performance cost that accounts for stealthy attacks. Theoretical analysis guarantees closed-loop boundedness despite the defender's information disadvantage, and simulations corroborate these results.

Nonequilibrium Dynamical Games: A Control Systems Perspective

Paper: [Annual Reviews in Control 2022]

Collaborators: Kyriakos G. Vamvoudakis, Aris Kanellopoulos, Nick-Marios T. Kokolakis

Dynamical games model strategic interactions between agents in changing environments, which is increasingly important as autonomous systems become more prevalent in society. This work applies bounded rationality concepts from economics—specifically level-k thinking and cognitive hierarchy—to predict nonequilibrium solutions in differential games using control theory frameworks. The approach is demonstrated through UAV pursuit-evasion scenarios and extended to stochastic games, providing rigorous mathematical guarantees for predicting game outcomes in both competitive and cooperative settings.

Prescribed Performance Control for Discontinuous Output Reference Tracking

Paper: [IEEE TAC 2021]

Collaborators: George A. Rovithakis

This article presents a low-complexity prescribed performance tracking controller for uncertain multi-input multi-output nonlinear systems that guarantees tracking errors converge within fixed time with predefined maximum overshoot, even with discontinuous reference signals. The approach modifies standard prescribed performance control by incorporating impulsive dynamics, eliminating the need to predict future reference values or discontinuity timings.

Detection of Actuator Faults for Continuous-Time Systems with Intermittent State Feedback

Paper: [Systems & Control Letters 2021]

Collaborators: Kyriakos G. Vamvoudakis

We present a fault detection method for uncertain continuous-time systems with intermittent state measurements, identifying actuator faults that create state evolutions incompatible with the system model. A necessary and sufficient detection condition is derived and relaxed for computational efficiency through a boundary-value problem formulation.

AE8750/8751: Robotics Research Fundamentals

Spring 2023 – Georgia Institute of Technology

Instructed full-load graduate-level course covering the fundamentals of research in robotics, including literature review, experimental design, and technical communication.

AE8803: Optimization-Based Learning Control and Games

Fall 2022, Fall 2023 – Georgia Institute of Technology

Delivered guest lectures and assisted with an advanced graduate course on optimization, learning, and game theory in control systems.