Saurabh Gupta
University of Illinois Urbana-Champaign, IL·Computer Science·Early-career
When people walk through an unfamiliar house, they still have useful expectations. Car keys are probably near the entrance, a door swings a predictable way, and an oncoming person will pass on one side.
- Fields
Computer Science
Engineering
- Topics
robotics
robot navigation
visual navigation
- Grants
- 1 known grant, through May 2027
What they work on
- Design predictive models of people, places and objects that a robot needs for decisions, such as which side an oncoming person will pass on or where household items are likely to be kept.
- Identify large, diverse data sources and develop ways to generate supervision (training signals) from them, so that predictive models for navigation and object interaction can be learned at scale.
- Build hierarchical, modular policy architectures that keep standard motion planners and feedback controllers and add learned modules for speculative choices, such as where to explore or where to stand to open a drawer.
- Measure how much predictive reasoning helps by comparing the efficiency of navigation and mobile manipulation systems with and without the learned predictive models.
More about the research
Robots lack these everyday expectations about people, places and things, and writing them in by hand is very hard. This project asks how machine learning can teach robots to make such predictions from large, varied datasets, so they can move around and handle objects more efficiently in places they have never seen.
Why it matters
Robots that can anticipate where objects are likely to be, how doors and drawers move, and where people will walk could get around homes and other new settings more efficiently. The grant names assistive robots and telepresence systems as practical uses that could benefit. The models, datasets and systems built here are also meant to support follow-up work on how robots and people interact.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Shuran Song
Stanford University, CA·Computer Science·Early-career
Develop hierarchical task representations, built in hyperbolic space (a tree-like curved geometry), learned from large-scale unlabeled instructional videos to model the steps of everyday activities such as cooking and repair.
Ioannis Karamouzas
University of California-Riverside, CA·Computer Science·Early-career
Build a constraint-based simulation framework that treats muscles, tendons, skeleton and environmental contacts as one coupled implicit system, for stable, physiologically realistic simulation of contact-rich human motion.
Kris Hauser
University of Illinois Urbana-Champaign, IL·Computer Science
Develop graph neural networks (GNNs), models that treat soil as connected regions, to predict how sand or soil responds to robot actions and to plan digging that forms target shapes, updated from camera images and force sensing.
The school
University of Illinois Urbana-Champaign: 41 labs on PI Pages, $907M a year on research, 870 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.