Towards Large Behavior Models: Versatile and Dexterous Robots via Supervised Learning

Explore the cutting-edge advancements in machine learning and their transformative impact on robotics in this 55-minute talk by Ben Burchfiel, manager and co-lead for the Large Behavior Model project at Toyota Research Institute. Delve into the potential of creating general-purpose robots through scaled machine learning techniques, drawing parallels with recent breakthroughs in language and vision models. Discover how robots can now learn complex behaviors like egg-beating and clothes-folding with minimal training time and computational resources. Gain insights into the development of Large Behavior Models for robots, which aim to combine the flexibility of Large Language Models with the ability to manipulate the physical world. Learn about the speaker's background in 3D perception, learning from demonstration, and multimodal vision-language representations, and understand his current focus on creating general-purpose robots using data-driven methods with soft inductive biases.