Yanbang Li Chinese Name Yanbang Li English Name

Hello, I am currently researching at Tsinghua University(IIIS), advised by Prof. Mengdi Xu. Previously, I worked at the EIL Lab at United Imaging Healthcare. I hold a Master's degree from the Dyson School of Design Engineering at Imperial College London, and I am also an artist and designer with a Master of Arts from the Royal College of Art (RCA). My research interests include robot learning, agentic hierarchical control, human-robot interaction, and AI for creativity. In my view, AI is a science that rethinks how the world fundamentally works. That's why I explore it.

profile photo profile photo with both arms raised

My future research will focus on the following directions:

  1. Human-Centered Robotics
    Developing interactive robotic systems that meaningfully collaborate with people, as well as exploring new forms of robotic art and creative expression that have not previously existed in human society.
  2. Efficient Physical AI
    Building AI systems that understand space and time, predict motion, force, and acceleration, and execute actions through high-speed closed-loop control under the high-level planning of agents or advanced VLMs.
  3. Practical Applications of Embodied AI
    Developing AI-enhanced robotic systems for agriculture, construction, and manufacturing that can replace humans in hazardous or highly repetitive tasks and solve practical problems in specific real-world industries.

I am currently seeking a PhD position. If you are interested in my work, please contact me.

Email

Publication
Robotic Spatiotemporal Editing figure 1 Robotic Spatiotemporal Editing figure 2 Robotic Spatiotemporal Editing figure 3

Robotic Spatiotemporal Editing

Yanbang Li*, Bu He*, Yi Tian, Yu-Chiao Huang, Chenyu Zhang, Ruoqiu Chen, Yuanzhe Liu, Yuxiang Gao, Mengdi Xu†

Under review, 2026

Our method extends existing generative policies with a new capability: beyond simply completing tasks, they can now be precisely controlled in both space and time.

arXiv Project Website

project image

Robotic Visual Instruction

Yanbang Li, Ziyang Gong, Haoyang Li, Xiaoqi Huang, Haolan Kang, Guangping Bai, Xianzheng Ma

IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2025

We propose a novel paradigm for Human-Robot Interaction (HRI): using hand-drawn sketches for robotic instruction, as an alternative to native language instructions.

Projects (Designed and made by myself)

Papersenses

A novel approach of creating paper-based human-computer interfaces by drawing and origami, with computer vision and deep learning methods. (1)A game engine creating games and controllers by drawing on paper. (2)A series of paper-based sensors in physical computing system. (2021)

project image

Tornado-Potato Robot

Tornado Potato Climbing Robot is an novel solution for smoothy steady stair-climbing inspired by Tornado Potato. The robot utilizes Archimedean spiral as its locomotion mechanism, providing more stability and smoothness than both legged robots and tracked robots for stair climbing. (2020)

project image

Infinite 360° FOV Tracking AR Projection Mirror

A glasses-free Augmented Reality (AR) installation where the displayed content within a room space dynamically adapts to the user's viewpoint. By precisely manipulating mirror reflections, the system re-routes the projector's light path, significantly extending the AR projection's display range and facilitating dynamic interaction with the user's position and field of view. (2023)

project image

Sky Chair

A chair that if you do not sit on it, it will fly away. Indoors, the chair will stay on the ceiling; when you want to sit, you need to pull the chair down. Once the user leaves the chair, it will float back to the ceiling by itself. It redefines the spatial layout of furniture by freeing up floor space completely. (2019)

project image

Mimosa

Mimosa is a street installation design for london using real mimosa plant as a rain sensor. (This project uses capacitive technology to integrate mimosa plant leaves into the circuit, transforming them into tactile/rainwater capacitive sensors.) When the wheather get rainy, the water drop fall from the sky and touch the leaves of the real mimosa plant. Then the natural mimosa close their leaves while the robotic mimosa will open its leaves as a shelter for passengers. (2019)

网站模板基于 Jon Barron