Duncan Eddy is a Postdoctoral Fellow at the Stanford Institute for Human-Centered AI working with Professor Mykel Kochenderfer in the Department of Aeronautics and Astronautics. He previously completed his PhD in Aerospace Engineering at Stanford University, funded by the National Defense Science and Engineering Graduate Fellowship.
Duncan's research focuses on building safe, reliable automated decision-making systems for real-world operational environments. His approach combines classical optimization, decision-making theory, and reinforcement learning techniques. His current focus is on adaptive stress testing, which is the application of reinforcement learning to efficiently discover the most likely failures in complex real-world systems, then train AI models to become resilient to those failures. Duncan is currently leading research efforts to validate and enhance the robustness of large language models, robotic systems, and spacecraft.
Prior to joining HAI, Duncan was a Principal Applied Scientist at Amazon Web Services, where he worked on software services for large-scale distributed edge computing. He led the Constellation Operations and Space Safety Group at Project Kuiper and developed a fully-automated constellation operations system for Capella Space's synthetic aperture radar satellite imaging constellation.