Hi there 👋
I am Lijian, a Research Assistant at the shanghaitech MARS Lab, working with Prof. Sören Schwertfeger.
Currently, my work is focused on Dexterous Hand Manipulation, Diffusion Models, and Quadruped Control.
Before joining ShanghaiTech, I worked at the Industrial and Commercial Bank of China Software Development Center (Beijing). I obtained my Bachelor Degree in Computer Science and Technology from Tianjin University of Finance and Economics in 2020.
High-Quality, ROS Compatible Video Encoding and Decoding for High-Definition Datasets
Li, Jian | Xu, Bowen | Schwertfeger
ROBIO 2024
Utilizing Flowmatching and RL (Work in Progress)
- Our work aims to improve the generation effect of the 2024 ICRA article Language-Conditioned Affordance-Pose Detection in 3D Point Clouds.
- So far, our pose generation effect has far exceeded that of the article, with nearly 90% of the generated poses being correct.
 |
 |
- We will subsequently conduct real-world testing using the Fetch Robot
(Work in Progress)
- Responsible for the training of upright balance and walking gaits for the UNITree GO2.
- Currently I am focused on solving the upright gait and balance without commands.
-
- The project is based on the 2024 RAL paper "RESPRECT: Speeding-up Multi-fingered Grasping with Residual Reinforcement Learning."
- Integrated Gaussian Splatting code into the paper's code to train a Gaussian set prediction network, accelerating Gaussian inference during RL training.
- Developed mirroring code for actions and observations, deploying the right-hand grasping strategy to the left hand.
-
- Collected campus datasets using the Mapping Robot and Faro.
- Tested different encoders and their settings to find the optimal configuration in terms of compressing visual SLAM dataset results, size, quality, and encoding time.
- Developed scripts for compressing and playing back datasets, achieving a 55x compression ratio. Produced a ROBIO paper.
- Using the Faro dataset as ground truth, wrote a program to annotate dynamic objects by comparing point clouds from two datasets (still in progress).