Event
Ph.D. Dissertation Defense: Asim Zoulkarni
Thursday, July 30, 2026
3:00 p.m.
AVW 1146
Emily Irwin
301 405 0680
eirwin@umd.edu
ANNOUNCEMENT: Ph.D. Dissertation Defense
Name: Asim Zoulkarni
Committee:
Professor John S. Baras, Chair/Advisor
Professor Chrysa Papagianni
Professor Gang Qu
Professor Zhambyl Shaikhanov
Professor William C. Regli, Dean's Representative
Date/time: Thursday, July 30 at 3:00 PM
Location: AVW 1146
Title: Adaptive Decision Frameworks for Secure, Real-Time Cyber-Physical Systems
Abstract: Autonomous cyber-physical systems operate in safety-critical environments with partial observability, adversarial perturbations, and limited sensing, communication, and computational resources. This thesis develops adaptive decision frameworks that improve security, reliability, and performance across robotic sensing, strategic cyber defense, and networked resource allocation.
First, the thesis presents a multimodal anomaly detection framework for autonomous robots under adversarial perturbations. It combines LiDAR observations with wheel velocity measurements through variational latent modeling to identify inconsistencies between perception and internal motion. Evaluations on a ROS-based Clearpath Husky platform demonstrate reliable detection of man-in-the-middle and message-flooding attacks while satisfying real-time deployment constraints.
Second, the thesis develops game-theoretic defenses against advanced persistent threats targeting autonomous ground robots. A two-phase framework models network penetration and damage infliction. During penetration, the defender allocates limited protections across an attack tree under uncertain exploit outcomes and temporal resource constraints. If penetration succeeds, endpoint anomaly detection monitors malicious data injection at safety-critical nodes. Simulations based on a ROS 2-enabled Nova Carter case study in Isaac Sim show that these strategies delay attacker progress more effectively than heuristic defenses while balancing detection speed against false alarms.
Third, the framework is extended to zero-day vulnerabilities through a Bayesian extensive-form game that represents uncertainty over hidden attack paths. The resulting mixed strategy directs limited defenses toward regions associated with elevated risk and improves resource allocation relative to heuristic approaches.
Finally, the thesis applies adaptive decision-making to radio resource scheduling in programmable and disaggregated radio access networks, where network slices with different service requirements share infrastructure. A meta-learning controller based on online convex optimization adapts resource allocations to time-varying channel conditions without prior knowledge of future conditions. Implemented through extensions to EdgeRIC and evaluated in Keysight O-RAN Architect, the controller achieves sublinear dynamic regret under bounded variation assumptions and improves service-level performance over baseline policies.
