Sounghwan (Eric) Hwang

Sounghwan (Eric) Hwang

Ph.D. Candidate at Purdue University

Research Overview

I develop control and estimation methods for autonomous systems that must operate safely under cyberattacks, uncertainty, and limited energy resources.

My Ph.D. research at Purdue University connects resilient multi-agent autonomy, nonlinear UAV control, and battery-aware coordination.

View the research overview diagram Overview diagram of my research on multi-agent systems, UAV control, and energy-aware autonomy.
Conceptual illustration of red cyberattack signals corrupting one UAV sensor channel, with resilient control mitigating the attack within a cooperating four-UAV network.
Cyberattack resilience in networked systems

Research area 01

Resilient Multi-Agent Autonomy

I develop control and estimation methods for cooperating autonomous vehicles under sensor attacks. My work combines reactive attack mitigation with proactive, reachability-based assessment of potential collision risks.

Resilient control · Sensor fusion · Reachability

Selected paper IEEE L-CSS, 2025
Research details & publications: Resilient Multi-Agent Autonomy
Research Motivation:
Multi-agent systems (MASs) have gained significant attention for their ability to solve complex engineering problems through cooperative decision-making. The primary objective in MAS operation is to achieve consensus among agents (e.g., UAVs, robots, and autonomous vehicles) to accomplish collaborative tasks. For example, urban air mobility (UAM) operations can be modeled as a MAS, where aerial vehicles coordinate their actions—such as formation control or velocity-matching consensus—by exchanging information (e.g., position and velocity) with neighboring agents. However, this communication-based structure also makes MASs inherently vulnerable to malicious threats, including cyberattacks, disturbances, and system faults. To address these challenges, my research focuses on developing advanced control and resilience mechanisms that enhance the safety and security of MASs operating in adversarial environments.
Sensor attacks and their propagation through a network of cooperating UAVs.
Figure: System vulnerability (i.e., sensor disruptions and attack propagation via a network) of MASs under cyberattacks.
The following is the summary of our ongoing research:

Reactive Multi-Agent System Defense Strategy

Research Objective:
In this research topic, we aim to design resilient control and estimation algorithms that can directly mitigate the impact of adversities. To this end, we developed resilient sensor fusion and estimation algorithms that can filter out the malicious data/information embedded in measurement output. The following videos show the realistic UAM operation in Greater Atlanta with four AVs conducting reference tracking control with formation flight. The left video shows the off-nominal UAM operation with a high risk of collisions. However, the right video shows the resilient UAM operation using our proposed method with high-assured safety.

(a) Off-Nominal Flight Scenario
(b) Resilient Control under Sensor Fault
Publications:

Proactive Multi-Agent System Defense Strategy

Research Objective:
In this research topic, we focus on developing security metrics for multi-AVs that can measure the potential risk (e.g., collisions) by stealthy attacks. We specifically utilize an over-approximated ellipsoidal reachable set through the Lyapunov stability criterion. This reachable set (red-shaded ellipsoids) indicates the level of performance degradation (e.g., trajectory deviation) posed by attacks at certain future time steps. If there are overlaps between reachable sets, we can identify that associated AVs may have potential risks in terms of collisions during operation.

Publications:
Conceptual illustration of cyberattack-induced red ellipsoidal reachable sets predicting a possible collision with an obstacle, and a teal UAV trajectory proactively avoiding the hazard.
Reachability-based proactive collision avoidance

Research area 02

Safety-Critical UAV Control

I investigate how cyberattacks and model uncertainty affect UAV safety. This work connects reachability-based risk assessment with robust safety-filter design, including studies using Gazebo and PX4-ROS2.

Safety filters · Control barrier functions · Risk assessment

Selected paper IEEE L-CSS, 2026
Research details & publications: Safety-Critical UAV Control
Research Motivation:
Ensuring the operational safety of Unmanned Aerial Vehicles (UAVs) is a fundamental challenge in modern aerospace systems. UAVs are vulnerable to various adversarial threats, including disturbances, wind gusts, and cyberattacks. For instance, GPS sensors can be compromised by spoofing attacks, which may significantly degrade navigation and trajectory-tracking performance. My research aim to develop safety-critical control and assurance algorithms that enhance the safety, reliability, and resilience of UAV systems operating in adversarial environments.
Quadrotor control architecture showing GPS spoofing attack channels.
Figure: Control architecture of UAV and potential system vulnerability under cyberattacks.
The following is the summary of our ongoing research:

Risk Assessment for UAVs under GPS Spoofing Attacks

Research Objective:
In this research, we develop a model-based risk assessment methodology for quadrotor UAVs under GPS spoofing attacks. These attacks represent particularly severe cyber threats due to their covert nature, allowing them to significantly degrade system performance without triggering alarms. To address this challenge, we propose a reachability-based security metric to quantify the extent of performance degradation caused by potential stealthy attacks. This methodology can be applicable to UAV tracking control operations in urban-like environments, where GPS sensors are highly susceptible to compromise by attackers.

Reachability-based risk assessment results for a UAV under GPS spoofing. Additional UAV reachability and risk assessment results.
Publication:
Sounghwan Hwang*, Minhyun Cho, and Inseok Hwang, "LMI-Driven Reachability Analysis for Fuzzy Model-Based Nonlinear Systems Subject to Norm-Bounded Input Perturbations ", IEEE Control System Letters (L-CSS), vol. 10, pp. 55-60, Jan. 2026.

UAV Safety-Filter Design through Control Barrier Function

Research Objective:
This research propose a safety-critical controller for nonlinear affine systems under actuator cyberattacks and model uncertainties. The approach combines a robust sliding mode-based control barrier function (SM-CBF) to address model uncertainties and an LSTM-based attack detector to identify compromised actuator channels. Conventional CBF controllers are sensitive to model dynamics, leading to safety violations under uncertainties and attacks. The proposed SM-CBF ensures safety despite these challenges, while the LSTM-based detector swiftly identifies compromised inputs. The methodology's effectiveness is demonstrated through quadrotor UAV stabilization in a high-fidelity simulator with Gazebo and PX4-ROS2.

Quadrotor safety-filter architecture with an SM-CBF and LSTM attack detector.
Figure: Control architecture for quadrotor UAV using reconfigurable SM-CBF safety filter and LSTM-based attack detector.
Publication:
Sungsoo Kim, Minhyun Cho, Sounghwan Hwang*, and Inseok Hwang, "Safety-Critical Control for Nonlinear Affine System With Robustness and Attack Recovery", AIAA SciTech 2025: Cybersecurity, Orlando, Florida, Jan 2025.
Conceptual illustration of a quadrotor tracking a curved reference trajectory, with overlapping fuzzy membership functions.
From nonlinear dynamics to controller synthesis

Research area 03

Nonlinear UAV Tracking Control

I use fuzzy models to represent the nonlinear, coupled dynamics of quadrotor UAVs and design tracking controllers through linear matrix inequalities (LMIs). This framework links nonlinear system modeling with optimization-based controller synthesis.

Fuzzy models · LMI synthesis · Trajectory tracking

Selected paper AIAA SciTech, 2026
Research details & publications: Nonlinear UAV Tracking Control
Research Motivation:
Unmanned Aerial Vehicles (UAVs) exhibit highly nonlinear and strongly coupled dynamics among attitude, altitude, and position states, making precise control inherently challenging. Conventional controllers (e.g., PID controllers) often fail to handle such nonlinearities and time-varying flight conditions effectively. To address these challenges, a fuzzy model-based intelligent control framework offers an efficient solution by approximating complex dynamics with a set of local linear models and adaptively blending them through fuzzy inference. This approach enhances robustness and adaptability across various flight modes. The effectiveness of this human inference-inspired control strategy is demonstrated through tracking control of a quadrotor UAV.
Fuzzy model representation of quadrotor dynamics connected to LMI-based controller design.
Figure: Control system architecture of a fuzzy model-based approach.
(a) Quadrotor UAV tracking control in 3D representation
(b) 2D representation
Publications:
Sounghwan Hwang*, Minhyun Cho, and Inseok Hwang, "LMI-Driven Tracking Control of Fuzzy Nonlinear Cyber-Physical Systems: Application to Quadrotor UAVs in Urban-Like Environment", AIAA SciTech 2026: Intelligent Systems: Adaptive and Intelligent Control Systems, Orlando, Florida.
Conceptual illustration of lithium-ion battery measurements mapped from nonlinear dynamics to a lifted model and a charge-state estimate.
Koopman-based battery state estimation

Research area 04

Battery Modeling & State Estimation

I develop Koopman-based models and LMI-based observers to estimate the state of charge of lithium-ion batteries. My work connects data-driven representations of nonlinear battery dynamics with control-theoretic observer design.

Koopman operators · Data-driven modeling · SOC estimation

Selected paper ASME ALDSC, 2026
Research details & publications: Battery Modeling & State Estimation
Research Motivation:
The operation of modern engineering systems such as electric vehicles, renewable energy storage, and electrified aircraft rely heavily on lithium-ion batteries. Accurately modeling these battery dynamics is essential for improving their operational safety, reliability, and performance. However, many of these systems exhibit nonlinear behaviors that are difficult to capture using traditional model-based approaches, such as equivalent circuit model. My research is motivated by the need for simple and reliable methods to understand and estimate such complex dynamics. Particularly, I focus on developing data-driven modeling, control, and estimation methods for battery dynamics using the Koopman operator framework.
Battery data transformed into a Koopman model and an LMI-based observer for state-of-charge estimation.
Figure: Data-driven Koopman framework enabling stability-guaranteed SOC estimation via LMI-based observer design.
Publications:
  • Sounghwan Hwang*, Guanlin Wu, Minhyun Cho, Vishnu Vijay, and Inseok Hwang, "Koopman-Based State-of-Charge Observer Design for Lithium-Ion Batteries: An LMI-Based Framework", ASME Letters in Dynamic Systems and Control (ALDSC), vol. 6, no. 4, Oct. 2026.
  • Guanlin Wu, Sounghwan Hwang*, Zhou Chen, and Inseok Hwang, "Deep Koopman-Style Framework for Cross-Temperature State-of-Charge Estimation of Lithium-Ion Batteries", Modeling, Estimation and Control Conference (MECC 2026), Accepted on June, 2026.
Conceptual illustration of four networked quadrotor UAVs with different battery charge levels.
Cooperative control with battery constraints

Research area 05

Energy-Aware Multi-Agent Coordination

I design cooperative controllers that account for the limited battery resources of multi-agent systems. The framework combines state-of-charge-dependent interaction weights with LMI-based observer and controller synthesis to incorporate battery safety into consensus control.

Consensus control · Battery safety · Energy constraints

Selected paper IEEE L-CSS, 2026
Research details & publications: Energy-Aware Multi-Agent Coordination
Research Motivation:
Multi-agent systems, such as UAM fleets and multi-robot teams, are typically powered by onboard batteries with limited energy capacity. Since each agent must operate within its energy budget to sustain mission-critical tasks, explicitly incorporating energy constraints into the cooperative control design is essential. Ignoring such constraints can lead to premature power depletion, degraded performance, or even mission failure. My research aims to develop energy-aware cooperative control strategies that enable multi-agent systems to accomplish collaborative tasks while ensuring safe and efficient energy consumption across all agents.
Energy-aware consensus framework combining battery state of charge, interaction weights, and LMI-based control.
Figure: Energy-aware MAS consensus control framework — SOC-dependent interaction weights combined with LMI-based observer/controller synthesis to guarantee battery safety.
Publications:
Sounghwan Hwang*, Guanlin Wu, Minhyun Cho, and Inseok Hwang, "Energy-Aware Consensus Control for Multi-Agent Systems with Guaranteed Battery Safety via H-Infinity LMI Design," IEEE Control Systems Letters (L-CSS), vol. 10, pp. 2293-2298, Aug. 2026.