
Towards Trusted Human-Centric Robot Autonomy
Karen Leung, from the University of Washington, speaks in Stanford's AA289 Robotics and Autonomous Systems Seminar on how robots can safely and fluently interact with humans. She opens by examining how humans navigate crowded spaces without collision and introduces data-driven safety concepts built using control theoretic tools as inductive biases, aiming to capture how people actually reason about safety rather than relying on conservative worst-case margins. In the second half she turns to motion fluency, presenting research showing that legible, proactive robot behavior produces more prosocial interactions with people nearby. The talk argues that safety and fluency, often treated as competing goals in robot decision-making and control, can be made more compatible. Drawn from the Stanford Robotics and Autonomous Systems Seminar series, the lecture addresses applications spanning autonomous vehicles, household and warehouse robots, and search and rescue systems.