Hi there! Welcome to my home page. My name is Zhiyuan, though I usually go by Paul.
I am currently a CS Ph.D. student at UC Berkeley, advised by professor Sergey Levine. I am interested in reinforcement learning and building autonomous intelligent agents. Before this, I graduated magna cum laude from Brown University with an Sc.B. in Applied Math and Computer Science, where I was advised by professors George Konidaris and Michael Littman at Brown. Check out my research page for more information.
Outside of this, I like to jam on my guitar, play badminton, and watch standups and movies.
I love connecting with people. If you’d like to get in contact with me, please reach out at [zhiyuan_zhou at berkeley dot edu].
Recent News
last updated: Mar 2025
- Excited to announce AutoEval, a system that autonomously evaluates generalist robot policies in the real world. I’m super excited to open access this resource and see how it can help the community! Give it a try and evaluate your policy now :)
- New paper out on how to do RL fine-tuning without offline data retention! I’m excited to see how Warm-start RL (WSRL) can help scale up RL for real-world tasks. Update: WSRL is accepted at ICLR 2025!
- I am thrilled to announce the first big project I led in my Ph.D! Check out SOAR: Autonomous Improvement of Instruction Following Skills via Foundation Models. Catch SOAR at CoRL 2024 in Munich!
- I am going to present my undergrad work Tiered Reward: Designing Rewards for Specification and Fast Learning of Desired Behavior at the RL Conference. Come chat with me!
- Check out my undergrad thesis Policy Transfer in Lifelong Reinforcement Learning through Learning Generalizing Features.
- I’ve compiled some resources that helped me during my CS Ph.D. application, along with some of my own thoughts. Hopefully this will be helpful to some others out there!
- I’m honored that our paper Designing Rewards for Fast Learning is selected for oral presentation at RLDM2022! Check out the recording at 1:20:22.
- Our work Characterizing the Action-Generalization Gap in Deep Q-Learning is accepted into RLDM2022!