Danshi Li

I am a research engineer at Galbot, Led by Prof. He Wang, Dr. Zhizheng Zhang and Prof. Li Yi. I obtained Master's degree from New York University and Bachelor's at CUHK. Previously I interned at GalBot and the EPIC Lab at Peking University, advised by Prof. He Wang

Email  /  CV  /  Bio  /  Scholar  /  Twitter  /  Github

Research

I am interested in learning generalizable, dexterous interactions between robot and the real world. Up to now, my works have been concentrated on the methodology of training in simulation and transferring to real world.

I am co-leading the ongoing DexGraspNet-series [v1] [v2] of projects, which aims to generate large-scale dexterous hand grasping/manipulation data that potentially adhere to semantic conditions and/or human intentions. Feel free to contact me and discuss your demands on data, and I will try my best to fulfill them!

Recently I take great interest in the theory of geometric farics and its applications in RL, e.g. in dexterous hand manipulation and dexterous grasping . Looking for interested partners to work together!

We build a large-scale dataset of 429M dexterous grasping poses in 7500 cluttered scenes with benchmark simulation pipeline. Based on the abundance of data, we learn a generative dexterous grasp prediction model that efficiently leverage local geometric features. Our model achieves 90.7% success rate, and show strong robustness under downscaling of training dataset.

a framework for 6-DoF object suction grasp detection on production lines with generalizable NeRF reconstruction, with a focus on but not limited to transparent objects.

Work Experience

Education