Robotics / Physical AI

Chaoqi Liu

I love robots. I want to do strong science, and help people feel able to do it, too.

I am a 4th year undergraduate student at UIUC. I have worked with Yilun Du, Kris Hauser, Yunzhu Li, and Maxim Likhachev.

I write notes on robotics and AI. Feel free to reach out about research ideas, collaborations, or robotics in general.

Chaoqi Liu

Research interests

Right recipe to generalist robots.

I am interested in fundamental questions that can eventually bring us to generalist robots: the right action abstraction, the right scene representation, and the right architecture for connecting learning, planning, and reasoning in the physical world.

Oredered action representation

OAT studies action vocabularies that preserve causality, controllability, and flexible compute for robot policies.

Composable policies and skills

FDP and MMPC study how action distributions can be factored, combined, and adapted instead of learned as one opaque behavior.

Models and planners that stay useful

IxG, L-GBND, and HMA explore when learned models and classical structure make robot behavior more reliable.

Updates

News

  1. My favorite project so far, Ordered Action Tokenization (OAT), is accepted. I will present it at RSS 2026 in Sydney.
  2. Factorized Diffusion Policy (FDP) is accepted to RA-L. I will present it at IROS 2026 in Pittsburgh.
  3. Hybrid Diffusion Policy (HDP) is accepted to RA-L.
  4. Multi-Modal Policy Consensus (MMPC) is accepted to ICRA 2026. MMPC explicitly decomposes action space by sensory modality for flexible multimodal fusion. Also check out FDP, which does implicit factorization.
  5. Localized Graph-Based Neural Dynamics (L-GBND) is accepted. I will present it at ICRA 2026 in Vienna.

Research output

Publications

* indicates equal contribution. Representative papers are highlighted.

1 token
2 tokens
4 tokens
8 tokens

OAT: Ordered Action Tokenization

Chaoqi Liu, Xiaoshen Han, Jiawei Gao, Yue Zhao, Haonan Chen, Yilun Du

Robotics Science and Systems (RSS), 2026

Media coverage: RoboPapers, MarkTechPost

Flexible Multitask Learning with Factorized Diffusion Policy

Chaoqi Liu, Haonan Chen, Sigmund H. Høeg*, Shaoxiong Yao*, Yunzhu Li, Kris Hauser, Yilun Du

IEEE Robotics and Automation Letters (RA-L), 2026
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2026
Spotlight at RSS 2025 2nd Workshop on Semantic Reasoning and Goal Understanding in Robotics [Link]

Multi-Modal Manipulation via Multi-Modal Policy Consensus

Haonan Chen, Jiaming Xu*, Hongyu Chen*, Kaiwen Hong, Binghao Huang, Chaoqi Liu, Jiayuan Mao, Yunzhu Li, Yilun Du*, Katherine Driggs-Campbell*

IEEE International Conference on Robotics and Automation (ICRA), 2026

Media coverage: Video Friday

Hybrid Diffusion for Simultaneous Symbolic and Continuous Planning

Sigmund H. Høeg, Aksel Vaaler, Chaoqi Liu, Olav Egeland, Yilun Du

IEEE Robotics and Automation Letters (RA-L), 2026

Implicit Graph Search for Planning on Graphs of Convex Sets

Ramkumar Natarajan, Chaoqi Liu, Howie Choset, Maxim Likhachev

Robotics Science and Systems (RSS), 2024

Writing

Blog

All posts

How to design action tokens?

How should we design action tokens for autoregressive robot policies? OAT is an ordered, prefix-decodable action representation built to be compact, totally decodable, and easy to model.

Recognition

Awards and Honors

Bronze Tablet

Highest undergraduate honor at UIUC.

James N. Snyder Award for Scholastic Achievement

Awarded to 2 recipients annually.

I maintained a 4.0 GPA at UIUC; if interested, a complete coursework list is available here.