Hongchao Zhang

Postdoctoral Scholar at Vanderbilt University

Portrait of Hongchao Zhang

I am a Postdoctoral Scholar at the Institute for Convergent Software Integrated Systems (ICONS), Vanderbilt University. I work with Taylor T. Johnson, Meiyi Ma, and Ahmad F. Taha at the intersection of artificial intelligence (AI), control, formal methods, and cyber-physical systems.

Education

  • Ph.D. in Electrical Engineering, Washington University in St. Louis, 2025
    Advisor: Andrew Clark
    Thesis: Resilient Safe Control of Autonomous Systems
  • M.S. in Electrical and Computer Engineering, Worcester Polytechnic Institute, 2020
  • B.E. in Automation Engineering, Nanjing University of Aeronautics and Astronautics, 2018

Professional Experience

  • Postdoctoral Scholar, Institute for Convergent Software Integrated Systems (ICONS), Vanderbilt University, September 2025–present
  • Graduate Research Assistant, Washington University in St. Louis, August 2022–August 2025
  • Graduate Research Assistant, Worcester Polytechnic Institute, August 2019–August 2022

Research

In 1942, Isaac Asimov proposed that a robot may not injure a human being or, through inaction, allow a human being to come to harm. Eight decades later, this science fiction is rapidly becoming reality as autonomous vehicles, drones, and robots operate alongside people, and Asimov’s laws now inspire the central challenge of making autonomy safe. Turning his First Law into a working system requires more than avoiding dangerous actions: a vehicle must navigate, a robot must complete its task, and infrastructure must keep serving its community. Artificial intelligence (AI) can help these systems make decisions, but people also need to know when those decisions meet the requirements for dependable operation. My research aims to establish when they do, which aligns with America’s AI Action Plan (2025), calling for “secure-by-design, robust, and resilient AI systems” in safety-critical applications.

My research develops foundations and methods towards trustworthy Physical AI, with a focus on safety and resilience. I study how to establish when systems meet their requirements and how to guide their reasoning and behavior in the presence of errors, uncertainty, or disruptions. My long-term vision is to connect trustworthy AI reasoning with dependable operation in the physical world, enabling autonomous systems and critical infrastructure to fulfill their intended functions under changing and adverse conditions.

Physical AI research map connecting a sensing-decision-action loop and its adversarial and black-box challenges to resilient control, safe physical AI, anomaly detection, verification, and verified intelligent autonomy.
Figure 1. Physical AI systems operate through a closed loop of sensing, decision-making, and action, where adversarial interference and the opacity of AI decisions pose two key challenges. My research addresses them by integrating formal verification, control theory, and AI into verified intelligent autonomy.

Physical systems, including robots, power grids, and water networks, must meet operating requirements and safety regulations to remain functional and safe. These systems rely on sensors to observe their operating conditions and on algorithms to select actions intended to satisfy these requirements. As shown in Figure 1, two challenges arise. First, faulty or corrupted measurements can undermine safe operation. Second, AI decisions can be difficult to analyze and verify, making it harder to determine whether they meet the system’s requirements and how to adjust them when they do not. Together, these challenges pose key barriers to establishing the trustworthiness of physical AI systems. My research connects formal verification, control theory, and AI through three research areas.

My established work spans three connected areas:

  • Resilient and Safe Physical AI: neural control barrier functions, exact certificate verification, fault-tolerant control, and safety under sensor faults and attacks.
  • AI Verification and Interpretation: formal verification across neural architectures, neuro-symbolic reasoning, and methods for identifying and correcting errors in complex AI behavior.
  • Verified Intelligent Autonomy: integrated reasoning, verification, and protective control for autonomous systems operating under uncertainty and adversarial conditions.

My future research develops these foundations in two directions: Trustworthy AI Autonomy, which connects sensing evidence to safe and useful robot behavior, and Critical Infrastructure, which combines physical constraints, anomaly analysis, and formal verification for electric and water systems.