Plenary Sessions

Title: Scalable Robot Decision Making in the Open World: Planning and Plan Prediction

Abstract: A hallmark of intelligence is the ability to do the right thing in myriad unfamiliar situations. The classic model-based approach to robotics draws a sharp boundary between the closed, known world and the open, unknown world: robot performance is guaranteed only in known situations. Data-driven robot foundation models, with their vast common-sense knowledge, have blurred this boundary and dramatically expanded robot capabilities in the open world. In this talk, I will argue that the long-term goal of scalable, robust robot intelligence necessitates an integration of model-based planning and data-driven plan prediction. Scalability depends on the interplay, rather than the conflict, between structural priors and data coverage. I will illustrate the general thinking with our work on robots aiming to operate in the open world by generating and verifying hypotheses on the fly. I will also speculate how agentic programming will come into play in the near future.

Title: Physical AI: Learning Dexterous Robotic Skills

Keywords: Robotic manipulation; embodied foundation models; reinforcement learning

Abstract: Recent advances in dexterous manipulation, vision-language-action models, and large-scale human data are accelerating the development of robots capable of precise and adaptable real-world interaction. This talk reviews the progression from sensorimotor skill learning and sim-to-real reinforcement learning to data-driven manipulation using tactile demonstrations, teleoperation, multimodal datasets, and human video. It concludes by examining generalist embodied foundation models that integrate large-scale learning, shared autonomy, and human demonstrations to enable reliable dexterous manipulation across complex tasks.