Title: Computationally Efficient and Safe Control for Aerial Robotic Systems under Threat and Disturbance

Date: Tuesday, May 5th, 2026

Time: 10:00 AM 

Location: TSRB 423

Virtual Link: https://teams.microsoft.com/meet/277536990612040?p=To1pMC8JMAJuDVMn4S

Evanns G. Morales-Cuadrado

Robotics PhD Student

School of Electrical and Computer Engineering

Georgia Institute of Technology

Committee:

Dr. Samuel Coogan (Advisor) - School of Electrical and Computer Engineering & School of Civil and Environmental Engineering, Georgia Institute of Technology

Dr. Shreyas Kousik - School of Mechanical Engineering, Georgia Institute of Technology

Dr. Kyriakos Vamvoudakis - School of Aerospace Engineering, Georgia Institute of Technology

Dr. Yorai Wardi - School of Electrical and Computer Engineering, Georgia Institute of Technology

Dr. Saman Zonouz- School of Cybersecurity and Privacy  & Electrical and Computer Engineering, Georgia Institute of Technology

Abstract: 

We propose a comprehensive program for safe autonomy in robotic systems, with a particular focus on aerial platforms. This program is organized across three complementary layers: secure high-level control, safe trajectory planning under uncertainty, and computationally efficient low-level tracking. Preliminary work establishes a lightweight Newton–Raphson-flow tracking controller with demonstrated alpha-stability on a miniature blimp, achieving competitive performance against PX4, feedback linearization, and nonlinear model predictive control across quadrotor and lighter-than-air platforms. A second line of work develops RTD-RAX, a reachability-based trajectory design and runtime assurance framework that combines offline parameterized reachable sets for rapid candidate generation with online mixed-monotone reachability verification and trajectory repair, enabling safe planning under bounded disturbances while avoiding conservative planning. Building on this foundation, the proposed work extends RTD-RAX to learned Gaussian process disturbance models, transitions prior numerical ground-vehicle results to hardware, and proposes extensions to aerial platforms with hardware validation. We also propose improved trajectory repair strategies, and extending RTD-RAX from reach-avoid missions to more complex STL missions by learning which control parameters map onto given STL specifications. We further propose a standalone learning-accelerated signal temporal logic framework that reduces reliance on repeated mixed-integer linear program solves while preserving safety guarantees. Additionally, this work proposes cyber-physical inoculation, in which fault signatures and certified backup controllers expand a resilient safe operating envelope against adversarial attacks. Together, these proposed contributions advance fast, efficient, and resilient autonomy for safety-critical cyber-physical systems.