Bridging Safety and Learning in Human-Robot Interaction

Explore the intersection of safety and learning in human-robot interaction through this insightful lecture by Andrea Bajcsy from UC Berkeley. Delve into the challenges of ensuring safe interactions between robots and humans, particularly when robots are learning from and about people. Discover how treating robot learning algorithms as dynamical systems driven by human data can enhance safety in human-robot interaction. Learn about a Bayesian monitor that assesses the robot's learned model in real-time and how control-theoretic tools can quantify the robot's learning potential. Examine practical applications of these concepts in robot motion planning, allowing robots to adapt their behavior based on the trustworthiness of their learned human models. Engage with thought-provoking questions about the appropriate definition of safety in human-robot interactions and consider new perspectives on capturing subtle aspects of these interactions. Gain insights from Bajcsy's research, which combines methods from control theory and machine learning to develop frameworks and algorithms for human-robot interaction in various domains, including assistive robotic arms, quadrotors, and autonomous cars.