Hanchen Cui

I am a Ph.D. student in computer science at University of Minnesota Twin Cities, working with Prof. Karthik Desingh. I received my master's and bachelor's degree from Shanghai Jiao Tong University and Xidian University, respectively. Previously, I'm a researcher at Shanghai Qi Zhi Institute, where I work closely with Prof. Yang Gao at IIIS, Tsinghua University. Afterwards, I have a short visit at NUS, collaborating with Prof. Lin Shao

My research interest lies in robot learning and my long-term goal is to build a generalized robot that can accomplish long-horizon and complex tasks. To be mroe specific, I work on vision-language-action models, world models, representation learning, and legged locomotion.

Email: hanchen.cui147[at]gmail.com

I am actively seeking internship opportunities for Summer 2025. If you're interested in collaborating, feel free to reach out!

Email  /  Scholar

Publications

(representative papers are highlighted)

We propose a data-driven framework for fine-tuning locomotion policies, targeting these hard-to-simulate objectives. Our framework leverages real-world data to model these objectives and incorporates the learned model into simulation for policy improvement.

We propose a generalizable world model pre-tained by large scale and diverse video dataset and then fine-tune the world model to obtain an accurate dynamic function in a sample-efficient manner.

We present Semantic-Geometric Representation (SGR), a universal perception module for robotics that leverages the rich semantic information of large-scale pre-trained 2D models and inherits the merits of 3D spatial reasoning.

Research projects

(1) Build a fast-speed vision-language manipulation deployment system to accomplish a variety of tasks based on Franka Panda Robot.

(2) Enable robust remote control with low latency.

(3) Fine-tune the vision-language manipulation model(CLIPORT), achieving 90% success rate in real-world experiments.

(1) Establish a sim2real locomotion deployment codebase for legged robots.

(2) Build a real-world learning framework which collects data in real world and trains the policy model in a remote server.

(3) Design and deploy a simple yet effective reward function, where all the reward terms are acquired from the robot itself without extra sensors.

(1) Leverage the strong image understanding and high-level planning ability of large multimodal models(like GPT4-Vision), enabling legged robots to accomplish multi-step and complex tasks, such as sending packages, and going out to find an object.

(1) Build a foundation policy model from large-scale expert dataset like Open X-Embodiment.

(2) Extract and label action squences using large vision-language models.

(3) Training and inference use action-aware sequence data instead of MDP.

Modified version of template from here.