Xujie Song

M.Phil

Tsinghua University

Biography

I am an M.Phil graduated from the Intelligent Driving Lab (iDLab), Tsinghua University. I worked with Prof. Shengbo Li and Prof. Bo Cheng. Currently, my research covers neural network, reinforcement learning, and autonomous driving. I have been committed to building more intelligent and safer AI for the decision-making and control of automated vehicles, and to developing the next generation of diagram for neural network training.

Interests

Neural Network
Reinforcement Learning
Autonomous Driving
Intelligent Transportation

Education

Tsinghua University, 2021-2024
M.Phil in Vehicle Engineering
Minor in Big Data (certificate program)
Delft University of Technology, 2017-2021
Joint Education Program
Beijing Jiaotong University, 2017-2021
B.Eng in Traffic and Transportation
B.Eng in Computer Science and Technology (dual degree)

Experiences

Research Assistant

Tsinghua University

September 2021 – Present Beijing

Research Intern

QBosoN Ltd.

February 2024 – July 2024 Beijing

Research Intern

Alibaba Group Ltd.

July 2023 – September 2023 Beijing

Research Intern

MEGVII Technology Ltd.

March 2022 – January 2023 Beijing

Research Intern

ProbQuant Ltd.

January 2022 – February 2022 Beijing

Research Intern

SenseTime Group Inc.

June 2021 – August 2021 Shanghai

Featured Publications

Liangfa Chen*, Xujie Song*, Wenxuan Wang, Yinuo Wang, Liming Xiao, Junjie Zhao, Shengbo Eben Li, Fei Ma, Jingliang Duan

February, 2024 IEEE Transactions on Intelligent Transportation Systems (TITS), 2024 (under review)

A Smooth Reinforcement Learning Method for Trajectory Tracking and Collision Avoidance of Wheeled Vehicle

We propose a smooth policy network (LipsNeXt) and a smooth distributed soft actor-critic (DSAC-S) algorithm to achieve coordinated optimization of control precision and action smoothness.

Wenxuan Wang, Jingliang Duan, Xujie Song, Liming Xiao, Liangfa Chen, Shengbo Eben Li, Bo Cheng

December, 2023 IEEE Transactions on Intelligent Vehicles (TIV), 2024 (under review)

Smooth Filtering Neural Network for Reinforcement Learning

We propose a class of neural networks with low-pass filtering ability, named Smonet, to alleviate the non-smooth issue by learning a low-frequency representation within hidden layers.

Xujie Song, Tong Liu, Shengbo Eben Li, Jingliang Duan, Wenxuan Wang, Keqiang Li

November, 2023 Nature Communications (NC), 2023 (ready to submit)

Training Multi-layer Neural Networks on Ising Machine

We propose Ising learning algorithm, which is the first technique to train multi-layer feedforward neural networks on Ising machine (a dedicated quantum computer).

Yinuo Wang, Wenxuan Wang, Xujie Song, Tong Liu, Yuming Yin, Liangfa Chen, Likun Wang, Jingliang Duan, Shengbo Eben Li

August, 2023 Annual Conference on Neural Information Processing Systems (NeurIPS), 2024 (under review)

Smoothing Neural Network with Adaptive Liquid Time-constant for Reinforcement Learning Tasks

we propose a neural ODE network variant, called adaptive liquid-time constant (ALTC), to smooth out control actions. A mapping function is incorporated to estimate the speed of system dynamics.

Xujie Song, Jingliang Duan, Wenxuan Wang, Shengbo Eben Li, Chen Chen, Bo Cheng, Bo Zhang, Junqing Wei, Xiaoming Simon Wang

July, 2023 International Conference on Machine Learning (ICML), 2023

LipsNet: A Smooth and Robust Neural Network with Adaptive Lipschitz Constant for High Accuracy Optimal Control

We propose LipsNet, a smooth and robust neural network with adaptive Lipschitz constant, to deal with the action fluctuation problem in RL (reinforcement learning).

Xujie Song, Zexi Lin

February, 2023 arXiv preprint, 2023

Non-zero-sum Game Control for Multi-vehicle Driving via Reinforcement Learning

This paper constructs the multi-vehicle driving scenario as a non-zero-sum game and proposes a novel game control framework, which consider prediction, decision and control as a whole.

Publications

Liangfa Chen*, Xujie Song*, Wenxuan Wang, Yinuo Wang, Liming Xiao, Junjie Zhao, Shengbo Eben Li, Fei Ma, Jingliang Duan. A Smooth Reinforcement Learning Method for Trajectory Tracking and Collision Avoidance of Wheeled Vehicle. IEEE Transactions on Intelligent Transportation Systems (TITS), 2024 (under review).

Wenxuan Wang, Jingliang Duan, Xujie Song, Liming Xiao, Liangfa Chen, Shengbo Eben Li, Bo Cheng. Smooth Filtering Neural Network for Reinforcement Learning. IEEE Transactions on Intelligent Vehicles (TIV), 2024 (under review).

Xujie Song, Tong Liu, Shengbo Eben Li, Jingliang Duan, Wenxuan Wang, Keqiang Li. Training Multi-layer Neural Networks on Ising Machine. Nature Communications (NC), 2023 (ready to submit).

PDF Cite

Shengbo Eben Li, Xujie Song, Tong Liu, Jingliang Duan, Wenxuan Wang, Bo Cheng. A Method for Training Feedforward Neural Network on Ising Machine. Patent, No. 2023113797497, 2023 (under review).

Yinuo Wang, Wenxuan Wang, Xujie Song, Tong Liu, Yuming Yin, Liangfa Chen, Likun Wang, Jingliang Duan, Shengbo Eben Li. Smoothing Neural Network with Adaptive Liquid Time-constant for Reinforcement Learning Tasks. Annual Conference on Neural Information Processing Systems (NeurIPS), 2024 (under review).

Liangfa Chen*, Xujie Song*, Liming Xiao, Lulu Gao, Fawang Zhang, Shengbo Eben Li, Fei Ma, Jingliang Duan. Smooth Reinforcement Learning Based Trajectory Tracking for Articulated Vehicles. Journal of Harbin Institute of Technology, 2023.

Xujie Song, Jingliang Duan, Wenxuan Wang, Shengbo Eben Li, Chen Chen, Bo Cheng, Bo Zhang, Junqing Wei, Xiaoming Simon Wang. LipsNet: A Smooth and Robust Neural Network with Adaptive Lipschitz Constant for High Accuracy Optimal Control. International Conference on Machine Learning (ICML), 2023.

PDF Cite Code Poster Video

Xujie Song, Zexi Lin. Non-zero-sum Game Control for Multi-vehicle Driving via Reinforcement Learning. arXiv preprint, 2023.

PDF Code

Shengbo Eben Li, Wenxuan Wang, Zhilong Zheng, Congsheng Zhang, Yujie Yang, Yuhang Zhang, Jiaxin Gao, Jie Li, Xujie Song, Tong Liu. Offline Simulation System for the General Optimal Control Problem Solver (V2.0). Software Copyright, No. 2023SR0225863, 2023.

Code Docs

Xujie Song, Sunhao Dai, Chunyu Lin, Shutao Zhan, Yao Zhao. LiDAR Data Enhancement via Pseudo-LiDAR Point Cloud Correction. Journal of image and Graphics (JIG), 2021.

PDF Code

Chunyu Lin, Xujie Song, Sunhao Dai, Shutao Zhan. A Method and Device for 3D Point Cloud Densification Based on Stereo RGB Images. Patent, No. CN111612728B, 2020.