Computer Vision Seminar: Dr. Ankit Goyal, "Towards Robotic Foundation Models via Joint Reasoning of Architecture and Data."."
- Event Type
- Seminar/Symposium
- Sponsor
- Illinois Computer Vision Group
- Virtual
- Join online
- Date
- Dec 11, 2025 4:00 pm
- Speaker
- Dr. Ankit Goyal
- Contact
- Allison Mette
- agk@illinois.edu
- Views
- 4
- Originating Calendar
- Siebel School Speakers Calendar
Abstract: Generalist robots that can perform many tasks in unstructured environments are a long-term goal for AI and robotics. Foundation Models (FMs) have already transformed language and vision by combining massive datasets with architectures that are well matched to those datasets. Robotics has not seen the same leap in generalization, not because we lack neural network capacity, but because we lack an “internet for robots”: a scalable, uniform source of data with a standard input-output interface. Robot data is fragmented across heterogeneous sensors, action spaces, and tasks, and cannot be trivially extracted from in-the-wild scalable sources. My research agenda starts from a simple premise: in robotics, we cannot just scale models; we must jointly reason about the data we use and the architectures that learn from it. My work demonstrates this through three approaches that each match architectural design to data characteristics. By leveraging high-quality 3D data, I build efficient, precise policies through 3D- aware architectures. By leveraging cheap data sources like internet video, I broaden generalization through hierarchical interfaces. By working with large 2D robot datasets, I show that the simplest architectures aligning with pre-trained FMs often outperform complex alternatives.
Bio: Dr. Ankit Goyal is a Research Scientist at the NVIDIA Robotics Research Lab. His research focuses on foundation models for robotics and exploring the connection between 3D vision and robotics. He completed his Ph.D. in Computer Science at Princeton University and his M.S. in Computer Science and Engineering from the University of Michigan. He also holds a B.Tech. in Electrical Engineering from IIT Kanpur. Dr. Goyal is a recipient of many awards, including the RSS Pioneers Award , the Qualcomm Innovation Fellowship , and the NeurIPS Scholar Award.