Chengxing Lin 林程星

Embodied AI / 3D Vision

Chengxing Lin

About

I am a master's student in Computer Technology at the University of Electronic Science and Technology of China (UESTC), advised by Prof. Wen Li. I received my bachelor's degree in Computer Science and Technology from Guangdong University of Technology (GDUT).

I work on embodied AI and 3D vision, with a focus on vision-and-language navigation, spatial understanding, test-time training, and point cloud representation learning.

Research Interests

Publications

  1. Deformation-based In-Context Learning for Point Cloud Understanding

    Chengxing Lin, Jinhong Deng, Yinjie Lei, Wen Li

    CVPR 2026 First author PDF

  2. Resource-Efficient Masked Autoencoding Framework for 3D Representation Learning

    Jian Zhu, Jiale Zhao, Chengxing Lin, Luan Lyu, Ruichu Cai, Xiaohua Ren, Enhua Wu

    Neural Networks, 2026 Corresponding author

  3. PatchMixing Masked Autoencoders for 3D Point Cloud Self-Supervised Learning

    Chengxing Lin, Wenju Xu, Jian Zhu, Yongwei Nie, Ruichu Cai, Xuemiao Xu

    IEEE TCSVT, 2024 First author

Manuscripts

  1. TETRIS: Self-Supervised Task Representation Learning for Point In-Context Learning

    submitting

  2. NavTTT: Adaptive Parametric Memory via Hierarchical Test-Time Training for Vision-and-Language Navigation

    submitting

  3. GuidedNav: Shaping Vision-Language Navigation Representations through Pre-Action Attention and Spatial Guidance

    submitting

Experience

Agibot / VLN Algorithm Intern

2026.04 – Present

Agibot Nav core contributor. Built the SFT training framework for discrete-action and pixel-goal navigation, and led model training and iteration. Designed simulation and real-world data collection workflows; explored DAgger, spatial guidance, adaptive memory, and reinforcement learning for navigation.

Developed a real-robot, language-directed object navigation demo integrating target localization, continuous tracking, and closed-loop visual servoing.

Pony.ai / Perception Algorithm Intern

2024.10 – 2025.04

Worked on lightweight visual model pretraining and knowledge distillation for autonomous driving. Built a ten-million-scale small-object dataset and an offline feature extraction and distillation pipeline, transferring CLIP and Qwen2-VL visual representations to compact models for pedestrian and vehicle classification.

Education