Research

My research focuses on advancing the test and evaluation (T&E) methodologies for autonomous systems, with a particular emphasis on safety-critical applications such as autonomous aerial refueling and naval vessel identification. I have developed transferable training frameworks that enable deep neural networks (DNNs) to generalize across diverse operational scenarios, significantly reducing the costs and time associated with data collection. By leveraging synthetic and fabricated datasets, I introduced cost-efficient certification processes that maintain high performance in real-world conditions. My work has addressed critical issues such as environmental variability—quantifying the impact of lighting, noise, and occlusion on DNN reliability—and set new benchmarks in precision certification through advanced instance segmentation techniques. Additionally, I emphasize standardized metrics like precision, recall, and intersection-over-union to create robust and defensible T&E frameworks. These efforts, demonstrated through collaborative research with midshipmen and faculty, bridge the gap between state-of-the-art AI methodologies and practical, mission-critical autonomous systems, ensuring their safety, reliability, and operational efficiency in complex environments.

In recent years, data science and machine learning (ML) have emerged as transformative tools, often replacing or complementing traditional control algorithms to enable the autonomous operation of unmanned vehicles in uncertain and dynamic environments. However, one of the most pressing challenges lies in perception, where AI systems must accurately interpret sensor data to make real-time decisions in complex and unpredictable settings. Perception challenges, such as object detection, environmental understanding, and adaptation to adverse conditions, directly impact the reliability and safety of autonomous systems. While AI research has predominantly focused on single-vehicle navigation, significant hurdles remain in coordinating multiple vehicles. These include addressing safety concerns like collision avoidance and the operational costs of equipping each vehicle with advanced GPUs and high-resolution sensors.

My research seeks to tackle these challenges by leveraging bio-inspired control strategies, where only a subset of vehicles employs AI-driven schemes, such as neural networks or reinforcement learning, for advanced perception and decision-making, while others rely on simpler local interaction rules for alignment and connectivity. Moreover, my work delves into emerging issues in ML and AI, including ensuring robustness under adversarial conditions, improving model generalizability across diverse environments, and mitigating biases in decision-making processes. By addressing both perception-related challenges and the broader limitations of AI, my research aims to enhance the reliability, scalability, and operational efficiency of autonomous systems in real-world applications.

The self-organized and coordinated maneuvers observed in collective behavior offer transformative insights for advancing science and engineering, particularly in robotics and autonomous distributed systems. These natural behaviors have inspired bio-inspired algorithms applicable to critical technical areas, including synchronization, neural network architectures, and robust communication protocols. My research focuses on developing distributed, bio-inspired intelligent algorithms with direct applications in robotics and networked systems, such as coordinating fleets of unmanned vehicles and designing autonomous distributed systems capable of operating in dynamic and uncertain environments. A central focus of my work is addressing uncertainty—whether stemming from environmental variability, incomplete data, or communication delays—to enhance system robustness and reliability.

In addition, I am advancing research in decentralized and cooperative control systems, autonomous collective dynamics, and self-reconfigurable modular robotics. These systems are designed to maintain optimal performance under uncertain and dynamic conditions, enabling practical applications such as search and rescue missions, environmental monitoring and protection, and human-robot or animal-robot interactions. By incorporating principles of autonomy and distributed intelligence, this work aims to drive the development of next-generation robotics systems capable of adaptive, efficient, and scalable operation in real-world scenarios, such as interconnected communities and smart cities.

Several natural and physical systems can be modeled as networks of coupled dynamical systems, in which the nodes correspond to individuals and the edges represent the interactions among them. For control applications, it is required that such networked dynamical systems synchronize towards a desired and well known state such as leader’s trajectories. The extended network composed of leaders and followers can be assimilated to an augmented stochastic network in which with nodes are intermittently connected in time. Analysis of synchronization over such networks is attractive for its potential applications in science and engineering. I want to develop a class of control strategies that can be implemented on such time varying networks and I seek to establish analytical and experimental tools to investigate their effectiveness. My approach builds on elements of stochastic stability theory, consensus protocols, and information theory, contributing to the development of tools that can be expanded to study wireless communication in dynamic sensor networks 

Many natural and physical systems can be effectively modeled as networks of coupled dynamical systems, where the nodes represent individual entities and the edges capture their interactions. In control applications, it is critical for such networked dynamical systems to achieve synchronization toward a desired state, such as the trajectory of a leader. When incorporating leaders and followers, the resulting network can be interpreted as an augmented stochastic network with time-varying, intermittently connected nodes. Analyzing synchronization within such dynamic networks is particularly compelling due to its broad applications in science and engineering.

My research aims to develop a class of control strategies tailored for time-varying networks, leveraging stochastic network dynamics. This involves establishing analytical and experimental frameworks to evaluate their effectiveness. By building on principles of stochastic stability theory, consensus protocols, and information theory, my approach seeks to advance the understanding of synchronization mechanisms in these systems. These tools will also have significant implications for studying wireless communication in dynamic sensor networks, enabling robust and adaptive coordination in environments characterized by uncertainty and temporal variability.