The Foundation Trilogy – Improving proficiency for generalizable autonomy
Assistant Professor, Computer Science
Attendance Link: https://docs.google.com/forms/d/e/1FAIpQLSdzU9VehavfTCy6zZk7vXcXHEx-FtcWxWNqa091cx10YlUvSQ/closedform
Abstract
The promise of large-scale robot learning is to enable robot capabilities that generalize across embodiments and tasks. Recent advances have shown remarkable progress with an ever-growing complexity of tasks, embodiments, and environments. In contrast to hard-coded industrial robotic work, large-scale robot policies are often still brittle. They do not offer the high throughput desired for commercially viable real-world deployment. To replicate the success in language modelling, most current commercial efforts focus on increasing training data and scaling pretraining. This talk discusses what comes next. With advances in world modelling on the one hand and mid- / post- training techniques on the other, generalizable learning-based robot models are enabled to carry out tasks at a performance level beyond what they have seen in their training data. The talk will present some of our work towards that goal. It will cover training and test-time steering techniques to unlock the next level of robot capabilities.
Bio:
Igor Gilitschenski is an Assistant Professor of Computer Science at the University of Toronto, where he leads the Toronto Intelligent Systems Lab. He is also a faculty member at the Vector Institute. Previously, he was a (visiting) Research Scientist at the Toyota Research Institute. He was also a Research Scientist at MIT’s Computer Science and Artificial Intelligence Lab and the Distributed Robotics Lab (DRL). There, he was the technical lead of DRL’s autonomous driving research team. Dr. Gilitschenski joined MIT from the Autonomous Systems Lab of ETH Zurich, where he worked on robotic perception. He obtained his doctorate in Computer Science from the Karlsruhe Institute of Technology and a Diploma in Mathematics from the University of Stuttgart. His current research interests focus on leveraging large models to build and manipulate efficient scene representations and to use these representations for robot learning. He is the recipient of several best paper awards, including at the American Control Conference, the International Conference of Information Fusion, and the Robotics and Automation Letters.
Please visit https://stanfordasl.github.io/robotics_seminar/ for this quarter’s lineup of speakers. Although we encourage in-person attendance, recordings of talks will be posted also.