Towards Embodied Intelligence: From First Principles to General-Purpose Autonomy

This talk explores how to push the boundaries of robotic autonomy and mobility. We review the current state of adaptive legged locomotion and navigation policies, and discuss pathways toward policies that generalize beyond their developers’ original intentions. We examine how such policies can be trained across diverse modalities and data sources, spanning simulation and the real world. In addition, we explore the role of reasoning in handling out-of-distribution scenarios, motivating the development of new scene representations for mobile robots. Finally, we argue that grounding autonomy in first principles can enable more robust and general robotic systems, improving safety while accelerating their deployment in real-world environments.

Speaker: Jonas Frey, Stanford University and UC Berkeley