From Robot Modeling to Physical AI: Toward Interpretable and Safe Robot Intelligence (Talk by Henrik Krauss)

Henrik Krauss of the University of Tokyo gives a one-hour online talk tracing his research path from conventional robot modeling and sensing toward learning-based methods for intelligent robots. The talk covers soft robot sensing, physics-informed learning, and model predictive control before turning to his doctoral work on learning interpretable robot dynamics directly from video and using those learned models for control. A second thread looks at human visual attention and at artificial agents informed by that attention, and how the pieces fit together into a view of physical AI that stays safe and inspectable rather than opaque. The framing question is how to make robots capable without making their internal behaviour impossible for people to reason about. The speaker also reflects on studying and doing research in Germany, Hong Kong, and Japan, including international collaboration, finding research directions worth pursuing, and building an early academic career abroad. The session runs on Zoom and is open to anyone who registers.