Kyle Vedder researches self-improving robot learning at Physical Intelligence. He previously worked at Dyna Robotics and earned a PhD at the University of Pennsylvania, developing methods and benchmarks for learning scene flow from unlabeled point clouds.

Unclaimed profile

Kyle Vedder researches self-improving robot learning at Physical Intelligence. He previously worked at Dyna Robotics and earned a PhD at the University of Pennsylvania, developing methods and benchmarks for learning scene flow from unlabeled point clouds.

Career

Member of Technical Staff at Physical Intelligence from 2025 to Present
Member of Technical Staff at Dyna Robotics from 2024 to 2025
Research Intern at NVIDIA from 2024 to 2024
Research Intern at Argo AI from 2022 to 2022
Teaching Assistant, Applied Machine Learning at University of Pennsylvania from 2021 to 2021
Software Development Intern at Amazon Lab126 from 2019 to 2019
Research Assistant, AMRL at University of Massachusetts Amherst from 2016 to 2019
Software Engineering Intern at Google from 2017 to 2017
C++ Development Intern at Unidesk from 2015 to 2015
vedder.io

An overview of my PhD

Kyle explains his work on learning scene flow without labels, including ZeroFlow, evaluation metrics, EulerFlow and shared research tools.

Open ↗
vedder.io

The Case Against Human Hands

Kyle argues that simple, serviceable robot grippers and arms can be more practical than copying human hands, examining maintenance, mechanics and learning from human video.

Open ↗
vedder.io

What Is It Like to Do Mathematics Research with Frontier Agents?

Kyle describes exploring a graph-theory problem with frontier AI agents, distinguishing his own contributions from model-generated ideas and explaining the verification work needed to assess them.

Open ↗

Kyle's Collections

Kyle's Lists

In these lists

Cosigns Given