Pieter Abbeel: Paving the Path to Generalizable Robotics

Pieter Abbeel wears many hats: Professor at UC Berkeley, Director of the Berkeley Robot Learning Lab, Founder of three companies, podcast host, and investor. The common thread is that Professor Abbeel is passionate about AI and robotics. In this keynote presentation, he explores the possibility of training a large neural network to enable faster learning in robotics. Professor Abbeel discusses his lab’s approach to solving this problem and will cover how video prediction is an excellent proxy for generalizable robots, the relevant models and datasets useful for pre-training, how unsupervised learning can help robots learn from themselves; and the usefulness of a human-in-the-loop. He describes a four-step framework that might be able to lead, ultimately, to generalized robotics. Professor Abbeel is co-director of the Berkeley Artificial Intelligence (BAIR) Lab and founded Gradescope, which provides AI to help instructors with grading homework and exams, and Covariant, which provides AI for robotic automation of warehouses and factories. He is also a founding partner at AIX Ventures, a venture capital firm focused on AI start-ups, and is the host of The Robot Brains podcast, which explores what AI and robotics can do today and where they are headed.