Historical claims data for autonomous systems doesn't exist. We turn
simulation data into the loss estimates insurers need to price coverage.
A foundational textbook covering algorithms for validating the safety of autonomous and cyber-physical systems, including methods for falsification, failure probability estimation, reachability analysis, explainability, and runtime monitoring.
Introduces a diffusion model-based approach for generating realistic failure scenarios to evaluate the safety of autonomous systems under rare but critical operating conditions.
A Bayesian optimization framework for efficiently estimating failure probabilities of black-box safety-critical systems, requiring orders of magnitude fewer simulations than Monte Carlo methods.
Introduces state-dependent importance sampling proposals to dramatically improve the efficiency of failure probability estimation for black-box autonomous systems.
Examines the formal requirements and practical considerations for certifying machine learning components in safety-critical aerospace applications.
A comprehensive framework for assessing and quantifying risk in autonomous vehicle systems across sensing, planning, and control, developed collaboratively across Stanford research groups and Allstate.
Stanford CS PhD · Previously at MIT Lincoln Laboratory, Xwing (now part of Joby Aviation), and NASA Ames Research Center
Stanford Aero/Astro PhD · Lecturer for Stanford's Validation of Safety-Critical Systems, previously at Reliable Robotics, MIT Lincoln Laboratory, and NASA
Stanford GSB Sloan Fellow & Actuary · Previously led insurance M&A at Dai-ichi Life Holdings
Stanford Aero/Astro PhD · Previously at Motional, The Aerospace Corporation, and MIT Lincoln Laboratory
Stanford University · Associate Professor of Aeronautics and Astronautics, Co-Director of the Stanford Center for AI Safety