2026 ACTION Annual Review and Knowledge Expo

Knowledge Expo Talks

Certified Learning, Control, and Diagnosis for Real-Time Trustworthy Robotics

Glen Chou
Professor, College of Computing
Georgia Institute of Technology

Abstract: Robots must make decisions under uncertainty, limited information, and the ever-present possibility of faults or adversaries. In critical domains, controllers must not only perform well, but also robustly satisfy constraints under perturbations. Formal reachability-based verification offers a principled way to provide such assurances, but applying these tools to robots is difficult: their dynamics are high-dimensional, AI-based components are difficult to certify, and existing verification methods are often too slow or conservative for online use.

In this talk, I will argue that trustworthy robotic autonomy requires moving beyond post-hoc verification toward the scalable co-design of controllers, learned models, and safety certificates. I will first present fast, differentiable reachability tools for neural planners and controllers, enabling certified training and online planning with worst-case robustness guarantees under bounded perturbations. I will then show how these reachability primitives support real-time active fault diagnosis, allowing robots to synthesize safe control policies that deliberately elicit informative observations to disambiguate actuator and sensor failures while maintaining safety. Finally, I will discuss how reachability can be informed by data-driven uncertainty sets to enable probabilistically safe control with learned latent world models. I will close with perspectives on future directions and opportunities in certifiable robotic autonomy.

Glen Chou is is an assistant professor at Georgia Tech in the College of Computing, within the School of Cybersecurity & Privacy (SCP), and in the College of Engineering, within the School of Aerospace Engineering (AE). He joined Georgia Tech in November 2024. Glen directs the Trustworthy Robotics Lab, which focuses on the design of algorithms that can enable general-purpose robots and autonomous systems to operate capably, safely, and securely, while remaining resilient to real-world failures and uncertainty. To achieve this, the lab leverages control theory and machine learning, while connecting to optimization, perception, formal methods, motion planning, human-robot interaction, and statistics. Glen received dual B.S. degrees in EECS and ME from UC Berkeley in 2017 and M.S. and Ph.D. degrees in ECE from the University of Michigan in 2019 and 2022, respectively. Prior to joining Georgia Tech in 2024, Glen spent two years as a postdoc at MIT CSAIL.

Dr. Chou directs the Trustworthy Robotics Lab, where he designs algorithms that can enable general-purpose robots and autonomous systems to operate capably, safely, and securely with humans, while remaining resilient to real-world failures and uncertainty. To achieve this, his group leverages control and machine learning, while connecting to optimization, perception, formal methods, planning, human-robot interaction, and statistics. Dr. Chou believes strongly in validating that the theoretical guarantees of his algorithms translate to the real world when deployed on hardware. He is interested in a broad range of applications, including robotic manipulation, vision-based navigation, aerospace autonomy, and the control of large-scale cyber-physical systems.

Offensive Informs Defensive: Building the Next Generation of Autonomous Defense

Kerri Prinos
Senior Research Staff Member
Horizon3.ai

Abstract: Modern cyber operations are being transformed by AI-enabled attackers, necessitating a fundamental shift in how we conceptualize adversarial engagement and defensive response. At Horizon3.ai, our team is building a learning loop between attack and defense to power the next generation of AI defenders. In this talk, we will present the results of our recent study to understand the AI attacker response to deception-based defenses, a novel tool-mediated agentic architecture for stable, autonomous, decision making in high-stakes, adversarial environments, and an evaluation framework designed to close the sim-to-real gap for evaluating autonomous defense agents in enterprise environments. Related paper: https://arxiv.org/abs/2605.03034

Kerri Prinos is a Senior Research Staff Member at Horizon3.ai, where she leads applied AI research for adversarial agentic cybersecurity systems. Her research interests include agentic AI, control theory, game theory, adversarial ML, algorithm development, low-SWaP, embedded AI, and applied mathematics. She has mentored over 30 undergraduate and graduate students across SOCOM Ignite, Joint All-Domain Strategist (Air University), summer research, and co-op programs. Prior to joining Horizon3.ai, she worked at MIT Lincoln Laboratory, researching machine learning and algorithm development across a wide variety of applications, including intelligence, surveillance, and reconnaissance; sensor fusion; autonomous systems; low-SWaP ultrasound technology; AI-driven offensive cybersecurity; and missile defense technology. Prinos was recognized as a two-time MIT Lincoln Laboratory Team Award winner and Technology Office Challenge Finalist for contributions to the Advanced Aerial Network Demonstration Team, student mentorship, and the multi-disciplinary low-SWaP AI-enabled ultrasound initiative. She holds an M.S. and B.S. in electrical engineering with a second major in systems science and engineering from Washington University in St. Louis and a B.S. in biology with a minor in applied mathematics from Davidson College.