Lingyu Xiao

Lingyu Xiao (Sephirex)

I am currently a MSc student at the School of Automation, Southeast University, China. Expected to graduate in 2025.

My research interests lie in:

► Perception & decision-making in autonomous driving and robotics.

► Multi-modal machine learning (e.g., vision and language).

Lots of my projects during undergraduate are focus on embedded systems in robotics (ROS, gazebo, etc.)

lyhsiao@seu.edu.cn / Google Scholar / Github

News

[Mar/2025] One first-authored paper accepted by Pattern Recognition!
[Jan/2025] One first-authored paper accepted by ICRA!
[Oct/2023] Start my internship at Baidu, VIS, Shanghai.
[Oct/2023] Travel to ICCV 2023 in Paris, France.
[Jul/2023] One first-authored paper accepted by ICCV!
[Sep/2022] Start my MSc at Southeast University, China.

Research

*Equal contribution. Some papers are highlighted.

Multi-modal Machine Learning

Revisiting MLLMs: An In-Depth Analysis of Image Classification Abilities
Huan Liu*, Lingyu Xiao*, Jiang-jiang Liu, Xiaofan Li, Ze Feng, Sen Yang, Jingdong Wang.
Preprint, Arxiv , 2024
Paper

Why some VLMs are bad at image classification while others are not?

Decision-Making in Autonomous Driving

Learning Multiple Probabilistic Decisions from Latent World Model in Autonomous Driving
Lingyu Xiao, Jiang-jiang Liu, Sen Yang, Xiaofan Li, Xiaoqing Ye, Wankou Yang, Jingdong Wang.
International Conference on Robotics and Automation (ICRA), 2025
Paper / Page / Code

Deriving decisions from an autoregressive latent world model through multiple probabilistic hypotheses. Expert-level performance on Waymax.

EasyChauffeur: A Baseline Advancing Simplicity and Efficiency on Waymax
Lingyu Xiao, Jiang-jiang Liu, Xiaoqing Ye, Wankou Yang, Jingdong Wang.
Preprint, Arxiv , 2024
Paper / Supp. Videos

Training strategies, data efficiency, and robust evaluation are equally important in learning based planner.

Perception in Autonomous Driving & Robotics

	Domain Adaptive Depth Completion via Spatial-Error Consistency Lingyu Xiao, Jinhui Wu, Junjie Hu, Ziyu Li, Wankou Yang. Pattern Recognition (PR), 2025 Addressesing the UDA problem in depth completion based on a domain/structural-agnostic property.
	ADNet: Lane Shape Prediction via Anchor Decomposition Lingyu Xiao, Xiang Li, Sen Yang, Wankou Yang. International Conference on Computer Vision (ICCV), 2023 Paper / Poster / Code Making the line anchor learnable in anchor-based lane detection methods. Achieving SOTA on VIL-100.

Projects

	Line-Following and Treasure-Hunting Robot RoboCup Competition, 2021 National 1st Prize in Robot Tourism Group Techniques: Electronic design, Mechanical Design, Control, Computer Vision Full Video The project requires robots to complete exploration and treasure-hunting tasks within 180 seconds, testing planning, adaptability, positioning, climbing, and obstacle-crossing abilities, while balancing speed and stability.
	Autonomous Driving Robot University Students Intelligent Car Race, 2021 National 3rd Prize in Creative Group Techniques: Computer Vision, Mechanical Design, Control Full Video / Technical Report (Chinese) The competition requires smart cars to follow the lane, identify specific markers, and complete tasks. Finishing is marked by crossing the finish line, and teams are ranked based on total score and time.
	Drone Landing with Visual Aid Course Project Techniques: SLAM, Computer Vision The drone needs to navigate through the unknown maze and land safely.

Honors

National Scholarship, Ministry of Education of P.R. China, October 2020
National 1st Prize, RoboCup Competition, Chinese Association of Automation, October 2021&2020
Natoinal 3rd Prize, University Students Intelligent Car Race, Chinese Association of Automation, 2021
4/4205, WAIC AD Multi-task Learning Challenge, BOSCH China, September 2022

Website Template