Explore new perspectives on leveraging foundation models for robot learning in this insightful conference talk by Mandi Zhao. Delve into two recent projects that address the challenge of incorporating high-level pre-trained capabilities into robotic systems for low-level embodied tasks. Discover RoCo, a project utilizing zero-shot Large Language Models (LLMs) as a communication tool for multi-robot collaboration. Learn about Real2Code, an innovative approach to the Real2Sim2Real problem for articulated objects that adapts both LLM and large pre-trained vision models. Gain valuable insights from Zhao's research at Stanford University and her experiences at Meta AI and Nvidia Seattle Robotics Lab. Understand the potential of data-driven approaches in enabling embodied systems to perceive, reason, and make sequential decisions in the real world.