Erin Zhang

Xiaoyu (Erin) Zhang

email github scholar
↓
about

I work at 1X Technologies, building vision-language-action models and vision-language models for humanoid robots. Lots of rapid prototyping, trying different things, and chasing signals to figure out what actually works on robots.

Before this, I was a graduate research assistant at the University of Illinois Urbana-Champaign, working with Saurabh Gupta on robotic manipulation and imitation learning. I completed my Master's in Computer Science at Carnegie Mellon University, co-advised by Oliver Kroemer and Yonatan Bisk. During undergrad (also at CMU), I worked with Andreas Pfenning on multi-species genome alignment.

I also spent two summers at Uber ATG, working on the Simulation and Prediction teams.

publications
2024
RSS

Diffusion Meets DAgger: Supercharging Eye-in-hand Imitation Learning

Uses diffusion models to synthesize out-of-distribution training data for imitation learning. DMD achieves 80% pushing success from just 8 expert demos, where behavior cloning manages 20%. Also works for stacking (92%), pouring (80%), and shirt hanging (90%).
2023
CoRL

Push Past Green: Learning to Look Behind Plant Foliage by Moving It

Self-supervised method for robots to reveal hidden space beneath plant foliage. Trains SRPNet to predict what space is uncovered by candidate actions, then chains actions to incrementally reveal more. Yes, it's robot gardening.
2020
Bioinformatics

HALPER: Regulatory Element Orthologs Across Species

Tool for constructing contiguous regulatory element orthologs from halLiftover outputs, enabling identification across hundreds of species. From my computational biology era.
teaching @ cmu
15-312 Foundations of Programming Languages
Spring 2020
15-150 Functional Programming
Spring 2018