About

I am a graduate student in Electrical Engineering and Computer Sciences (EECS) at UC Berkeley, with a focus on reinforcement learning and robotics. I am currently a Graduate Research Assistant at Berkeley Artificial Intelligence Research (BAIR), where I work with Professor Avideh Zakhor on robot learning for locomotion, manipulation, and real-world automation.

I received my B.A. in Mathematics and Computer Science from Berkeley and am now pursuing an M.S. in EECS. My background includes coursework and project work in computer vision, deep reinforcement learning, machine learning, robotics, and related areas. I am actively seeking research and industry opportunities in robotics, machine learning, and applied AI.

Experience

Graduate Research Assistant

Berkeley Artificial Intelligence Research (BAIR) · June 2025 – Present

Developing locomotion and manipulation policies for Unitree G1 humanoid robots with Professor Avideh Zakhor. Running experiments in Isaac Sim via Isaac Lab with PPO and imitation learning for whole-body control, with the goal of simulation-to-real deployment.

Undergraduate Research Assistant

Berkeley Artificial Intelligence Research (BAIR) · Jan 2024 – May 2025

Applied deep reinforcement learning to a real-world tracked robot for vacuuming and sealing applications. Developed ROS2 and Gazebo simulation pipelines and trained CNN-LSTM policies with PPO for autonomous navigation.

Teaching Assistant & Course Staff

UC Berkeley EECS · 2022 – 2025

Tutor for CS 182 (Deep Learning), reader for CS 189 (Machine Learning), and academic intern for CS 61A (Structure and Interpretation of Computer Programs). Led discussion sections, developed and graded assignments and exams, and supported hundreds of students across introductory programming, machine learning, and deep learning.

Research

Unitree G1 Humanoid Robot

Humanoid Loco-Manipulation

Berkeley Artificial Intelligence Research (BAIR) · 2025 – Present

Developing locomotion and manipulation policies for the Unitree G1 humanoid robot. Using Isaac Sim and Isaac Lab with PPO and imitation learning for whole-body control, with the goal of simulation-to-real deployment.

Tracked robot

Tracked Robot Control

Berkeley Artificial Intelligence Research (BAIR) · Jan 2024 – May 2025

Applied deep reinforcement learning to a real-world tracked robot to automate vacuuming and sealing American attics. ROS2, Gazebo, PPO, CNN-LSTM policy for autonomous navigation.

ICL architecture

Can Custom Models Learn In-Context? An Exploration of Hybrid Architecture Performance on In-Context Learning Tasks

R. Campbell, N. Lojo, K. Viswanadha, et al. · Nov 2024

Explores how hybrid architectures (GPT-2/LLaMa, LLaMa/Mamba) perform on in-context learning tasks. Proposes the "ICL regression score" and releases a typed, modular Python package.

ACL framework

Automatic Curriculum Learning with Gradient Reward Signals

R. Campbell, J. Yoon · Dec 2023

Uses gradient norm reward signals in automatic curriculum learning for DRL. Evaluated on PointMaze, AntMaze, and AdroitHandRelocate. Shows accelerated learning and improved generalization.

In-context learning of function classes (Garg et al.)

Can Transformers Learn Sequential Function Classes In Context?

R. Campbell, E. Guo, E. Hu, R. Vir, E. Hsiao · Dec 2023

Extends the in-context learning framework of Garg et al. to sequential function classes. We introduce a sliding-window sequential function class and show that transformers can in-context learn these tasks; we also study robustness under randomized labels.

Projects

Update rules: SGD, Muon, and Schatten norm

Interpolating Muon

Project on interpolating optimizers and the geometry of gradient updates. Analysis of matrix update rules and their associated norms: SGD (Euclidean), Muon (spectral norm), and a family of Schatten-(1+1/p) norms. Proposes a cheaper alternative to Muon that matches its empirical performance.

NeRF lego

CS 180: Computer Vision

Course project writeups: image processing, filters, face morphing, photo mosaics, diffusion models, neural radiance field.

Coup agent

Coup: RL agent for the card game

Reinforcement learning agent for Coup. Agent network with game state and history, action heads with masking for legal moves.

Sawyrter: Sawyer robot with colored blocks

Sawyrter

Robotics project using a Sawyer collaborative arm for object manipulation, with computer vision for detecting colored blocks and ArUco markers. Team project at Berkeley.

WhereToEat Chatbot interface

WhereToEat Chatbot

AI chatbot built with a team of 4 at the Cal Hacks AI Hackathon to help with restaurant selection. Next.js frontend (deployed on Vercel), Python backend using OpenAI and Google Maps APIs.

Chain-of-Thought autoregressive process

Sample Complexity for Agnostic Learning with Autoregressive Chain of Thought

Sample and computational complexity of learning prompt-to-answer mappings when the model generates a chain-of-thought and the final token is the answer. Inspired by Joshi et al. on learning with autoregressive chain-of-thought.

Condorcet consistent voting diagram

Algorithm(s) and Complexity for Condorcet Consistent Voting

Literature review with Amar Shah on complexities for Condorcet-consistent voting systems (Dodgson, Kemeny, Ranked Pairs) and applications to simple stable voting.

Music

I am a jazz musician who plays bass and piano. I write and record my own music, and I also perform with Frankly Martini, a 7-piece jazz fusion band I co-founded with fellow musicians from Berkeley. Our album Wakey Wakey is out now. I also play in The Fabric, a 4-piece indie rock band.

Check out my personal Spotify, Frankly Martini on Spotify and Instagram, and The Fabric on Instagram.

I've studied through UC Jazz with Frank Martin, Ted Moore, and Dan Zinn. I've also taken private lessons on bass with Daniel Lucca Parenti and on piano with Joan Cifarelli. I teach private lessons and lead my own ensemble in the UC Jazz program.

Contact

Reach out via the links below.

ryancampbell [at] berkeley [dot] edu