Brent Yi

Hello! I'm a final-year CS PhD student at UC Berkeley, where I'm advised by Prof. Angjoo Kanazawa and Yi Ma. I'm broadly interested in building practical, scalable machine learning systems for embodied agents.

I'm currently an intern at Amazon FAR. My PhD work is supported by the NSF Fellowship. Before the PhD, I worked with Jeannette Bohg during an MS at Stanford, and Pieter Abbeel during a BA at UC Berkeley.

GitHub / Twitter / Email

Research

My research is at the intersection of machine learning, computer vision, and robotics. These days, I've been thinking about three areas:

Expressive, scalable learning for action. Reinforcement learning-based control using flow matching^{1Flow Policy Gradients for Robot Control (2026),2Flow Matching Policy Gradients (2026)} and in-the-wild motion capture from humans^{4Estimating Body and Hand Motion in an Ego-sensed World
(2025)}.
Helping computers see in 3D. Encoding geometry in multiview transformers^{3Cameras as Relative Positional Encoding (2025)}, 3D reconstruction^{5Canonical Factors for Hybrid Neural Fields (2023),13Nerfstudio: Modular Framework for Neural Radiance Field
Development}.
Software for efficient research. I've open-sourced visualization^{7Viser: Imperative, Web-Based 3D Visualization for Python}, configuration^{8Tyro: Type-Safe CLI and Config Generation for Python}, and nonlinear optimization^{9jaxls: Sparse Nonlinear Least Squares in JAX,10pyroki: Modular Toolkit for Robot Kinematic Optimization} tools that I've built for my work. I've also contributed significantly to frameworks for reinforcement learning-based control^{11mjlab: Lightweight Framework for GPU-Accelerated Robot
Learning,12Holosoma: Humanoid Robotics Framework for RL Training and
Deployment} and novel view synthesis^{13Nerfstudio: Modular Framework for Neural Radiance Field
Development,14gsplat: Gaussian Splatting CUDA Kernels with Python
Bindings}.

Feel free to reach out if you're interested in chatting or collaborating!

Publications

Full list on Google Scholar!

[1]

Flow Policy Gradients for Robot Control

arXiv 2026

Website Paper

Brent Yi*, Hongsuk Choi*, Himanshu Gaurav Singh, Xiaoyu Huang, Takara E. Truong, Carmelo Sferrazza, Yi Ma, Rocky Duan, Pieter Abbeel, Guanya Shi, Karen Liu, Angjoo Kanazawa.

[2]

Flow Matching Policy Gradients

ICLR 2026

Website Paper Code

David McAllister*, Songwei Ge*, Brent Yi*, Chung Min Kim, Ethan Weber, Hongsuk Choi, Haiwen Feng, Angjoo Kanazawa.

[3]

Cameras as Relative Positional Encoding

NeurIPS 2025

Website Paper Code

Ruilong Li*, Brent Yi*, Junchen Liu*, Hang Gao, Yi Ma, Angjoo Kanazawa.

[4]

Estimating Body and Hand Motion in an Ego-sensed World

CVPR 2025 (Highlight)

Website Paper Code

Brent Yi, Vickie Ye, Maya Zheng, Yunqi Li, Lea Müller, Georgios Pavlakos, Yi Ma, Jitendra Malik, Angjoo Kanazawa.

[5]

Canonical Factors for Hybrid Neural Fields

ICCV 2023

Website Paper Code

Brent Yi, Weijia Zeng, Sam Buchanan, Yi Ma.

[6]

Quasi-direct Drive for Low-Cost Compliant Robotic Manipulation

ICRA 2019

Website Paper Code

David Gealy, Stephen McKinley, Brent Yi, Philipp Wu, Phillip Downey, Greg Balke, Allan Zhao, Menglong Guo, Rachel Thomasson, Anthony Sinclair, Peter Cuellar, Zoe McCarthy, Pieter Abbeel.

Software

It's important to me that research artifacts are easy to run, reproduce, and debug:

[7]

viser

Imperative, web-based 3D visualization for Python.

arXiv Docs Code

Brent Yi, Chung Min Kim, Justin Kerr, Gina Wu, Rebecca Feng, Anthony Zhang, Jonas Kulhanek, Hongsuk Choi, Yi Ma, Matthew Tancik, Angjoo Kanazawa.

[8]

tyro

Type-safe CLI and config generation for Python.

Docs Code

Brent Yi.

[9]

jaxls

Solve sparse, non-Euclidean, and constrained nonlinear least squares problems in JAX.

Docs Code

Brent Yi.

[10]

pyroki

Modular toolkit for robot kinematic optimization.

IROS 2025 Paper Docs Code

Chung Min Kim*, Brent Yi*, Hongsuk Choi, Yi Ma, Ken Goldberg, Angjoo Kanazawa.

[11]

mjlab

Lightweight framework for GPU-accelerated robot learning.

arXiv Docs Code

Kevin Zakka, Qiayuan Liao, Brent Yi, Louis Le Lay, Koushil Sreenath, Pieter Abbeel.

[12]

holosoma

Humanoid robotics framework for RL training, deployment, and motion retargeting.

Code

Pieter Abbeel, Juyue Chen, Rocky Duan, Alejandro Escontrela, Manan Gandhi, Samuel Gundry, Xiaoyu Huang, Angjoo Kanazawa, Tomasz Lewicki, Jiaman Li, Karen Liu, Clay Rosenthal, Younggyo Seo, Carlo Sferrazza, Guanya Shi, Linda Shih, Jonathan Tseng, Zhen Wu, Lujie Yang, Brent Yi, Yuanhang Zhang.

[13]

nerfstudio

Modular framework for neural radiance field development.

SIGGRAPH 2023 Paper Docs Code

Matthew Tancik*, Ethan Weber*, Evonne Ng*, Ruilong Li, Brent Yi, Justin Kerr, Terrance Wang, Alexander Kristoffersen, Jake Austin, Kamyar Salahi, Abhik Ahuja, David McAllister, Angjoo Kanazawa.

[14]

gsplat

Gaussian splatting CUDA kernels with Python bindings.

JMLR MLOSS 2025 Paper Code

Vickie Ye*, Ruilong Li*, Justin Kerr, Matias Turkulainen, Brent Yi, Zhuoyang Pan, Otto Seiskari, Jianbo Ye, Jeffrey Hu, Matthew Tancik, Angjoo Kanazawa.

Misc

I previously spent time on robot hardware design. Here's a miniature portfolio!