Generalist Robots
Chelsea Finn discusses the concept of training robots to have general purpose models that can predict the future based on their actions, allowing them to solve a variety of tasks. She highlights the benefits of self-supervised learning and the challenges of training video prediction models, planning procedures, and data acquisition. The potential of this approach is demonstrated through the ability to solve a wide range of tasks, even with objects the robot has never encountered before.In this clip
From this podcast
The Gradient
Chelsea Finn on Meta Learning & Model Based Reinforcement Learning
Related Questions