Shuran Song
Stanford University, CA·Computer Science·Early-career
Robots that can see, understand and act in messy everyday environments could help people at home and beyond. Most robot vision systems, however, only detect predefined objects in static images, and robots rarely anticipate what a person needs.
- Fields
Computer Science
Engineering
- Topics
robotics
robot manipulation
computer vision
- Grants
- 2 known grants, through Apr 2027
What they work on
- 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.
- Create methods for robots to predict and verify a person's intentions from vision and dialogue, and use them for intention-aware planning that delivers the right object at the right time.
- Train robot visual predictive models with self-supervision, using the robot's own manipulation actions and their observed effects as ground-truth labels.
- Use perception-model signals such as uncertainty, novelty and predictability as rewards to improve manipulation policies, and combine both models for efficient action planning.
More about the research
Both of the lab's projects aim to give robots richer understanding of scenes and tasks by learning from video, language and their own interactions with the world.
Why it matters
Robots that anticipate needs and hand over objects at the right time could help people with limited mobility, including older adults and veterans, cook, repair things and keep house more independently. Robots that learn about their surroundings by interacting with them could make service robots and field exploration robots more adaptable to unfamiliar, changing situations.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Changhyun Choi
University of Minnesota, MN·Computer Science·Early-career
Develop context-aware grasping algorithms that let robots consider clutter, reachability and collisions when picking objects from crowded scenes, using visual scene understanding and self-supervised learning.
Saurabh Gupta
University of Illinois Urbana-Champaign, IL·Computer Science·Early-career
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.
Sergei Nirenburg
Rensselaer Polytechnic Institute, NY·Computer Science
Assess the commercial translation potential of a cognitive-robotics technology through experiential learning and first-hand investigation of the industry ecosystem.
The school
Stanford University: 87 labs on PI Pages, $1.6B a year on research, 739 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.