I'm training robots to be robust and performant enough to deploy in human environments — through off-domain data, pretrained priors, world models, data curation, and test-time adaptation.
I was an undergraduate and now masters student in the WEIRD Lab at the University of Washington, advised by Abhishek Gupta. My journey loving robotics started in high school as captain of my FRC team.
I interned at the Amazon Robotics Innovation Lab on a prototype package-manipulation robot called Calisto, and was a predoctoral instructor at UW for CSE 311. Outside research, I care a lot about teaching — I've been an instructor or TA for nearly all of my time at UW.
I'm interested in robotics, reinforcement learning, and world modeling.
Jacob Berg, Chuning Zhu, Yanda Bao, Ishan Durugkar, Abhishek Gupta
Semantic World Models, unlike traditional world models, answer questions about the future given current observations and a sequence of actions.
Marius Memmel*, Jacob Berg*, Bingqing Chen, Abhishek Gupta†, Jonathan Francis†
STRAP uses dynamic time warping and pretrained visual foundation models to retrieve sub-trajectories that augment policy learning.
Jianlan Luo, Zheyuan Hu, Charles Xu, You Liang Tan, Jacob Berg, Archit Sharma, Stefan Schaal, Chelsea Finn, Abhishek Gupta, Sergey Levine
An open-source framework that helps facilitate wider usage of RL algorithms and ease deployment to real-world robots.
Daniel Yang, Davin Tjia, Jacob Berg, Dima Damen, Pulkit Agrawal, Abhishek Gupta
Effective shaped rewards can be created from “progress” toward task completion — works in sim, real, and on web-scale video.
I've been a 8x TA for CSE 311, as well as CSE 142 and CSE 579. In addition, I've been the instructor of record for the following courses:
Introducing students to the shared language and problem-solving toolkit of computer science — the foundation for tackling difficult, real-world computing problems.