Fan Lu

HelloWorld: Towards Practical Applications of Generative Driving World Models

HelloWorld Team (Fan Lu: first author)

Technical Report, 2026

A unified driving world model combining controllable multi-view video generation and LiDAR synthesis for scene editing, training data generation, and closed-loop simulation.

Urban Architect: Steerable 3D Driving Scene Generation with Layout Prior

Fan Lu^*, Kwan-Yee Lin^*, Yan Xu, Hongsheng Li, Guang Chen^†, Changjun Jiang

IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2026

Steerable, large-scale 3D driving scene generation from text and compositional layouts, with control over scene geometry, appearance, and editing.

Efficient Trajectory-Conditioned Text-to-4D Gaussian Splatting

Lin Shao, Fan Lu^†, Haiyun Wei, Sanqing Qu, Alois Knoll, Guang Chen^†

IEEE International Conference on Robotics and Automation (ICRA), 2026

Efficient generation of dynamic 4D assets along specified trajectories, combining Gaussian splatting with control-node deformation to preserve motion quality.

MGS-Track: Monocular 6DoF Pose Tracking via Masked 3D Prior and Online Gaussian Splatting

Zhiyuan Chen, Fan Lu^†, Guo Yu, Sanqing Qu, Ya Wu, Yuan Huang, Alois Knoll, Guang Chen^†

IEEE International Conference on Robotics and Automation (ICRA), 2026

Joint 6DoF object tracking and reconstruction from monocular video, using mask-aware geometric priors and online Gaussian splatting for robotic manipulation scenarios.

Signal Structure-Aware Gaussian Splatting for Large-Scale Scene Reconstruction

Weiyi Xue, Fan Lu^†, Chi Zhang, Tianhang Wang, Sanqing Qu, Zehan Zheng, Boyuan Zheng, Junqiao Zhao, Guang Chen^†

International Conference on Learning Representations (ICLR), 2026

Frequency-aware scheduling aligns image supervision with Gaussian densification, improving reconstruction quality and efficiency in large-scale scenes.

Multimodal LiDAR-Camera Novel View Synthesis with Unified Pose-free Neural Fields

Weiyi Xue, Fan Lu^†, Yunwei Zhu, Zehan Zheng, Haiyun Wei, Sanqing Qu, Jiangtong Li, Ya Wu, Guang Chen^†

Advances in Neural Information Processing Systems (NeurIPS), 2025

Unified LiDAR-camera neural fields jointly optimize sensor poses and scene representations, enabling multimodal novel view synthesis without accurate input poses.

An Online-Training-Free Adaptor for Open Heterogeneous Collaborative Perception via Diffusion Model

Tianhang Wang^*, Fan Lu^*, Sanqing Qu^†, Bin Li, Ya Wu, Hu Cao, Alois Knoll, Guang Chen^†

IEEE Transactions on Circuits and Systems for Video Technology (TCSVT) · Accepted 2025; Vol. 36, 2026

DiffAlign uses a pretrained diffusion model to align features across heterogeneous agents, enabling collaborative perception without online training or access to model internals.

Post-Refining LiDAR Point Cloud Registration via Diffusion-Based Correspondence Refinement

Fan Lu, Tianhang Wang, Guo Yu, Bin Li^†, Qiao Lin, Alois Knoll, Zhijun Li, Guang Chen

IEEE Transactions on Automation Science and Engineering (T-ASE) · Accepted 2025; Vol. 23, 2026

DCR refines noisy point correspondences with a conditional diffusion model, improving the accuracy of both learned and traditional LiDAR registration methods.

Urban Radiance Field Representation with Deformable Neural Mesh Primitives

Fan Lu^*, Yan Xu^*, Guang Chen^†, Hongsheng Li, Kwan-Yee Lin^†, Changjun Jiang

International Conference on Computer Vision (ICCV), 2023

This paper proposes a novel neural rendering framework for urban scenes. We represent the urban scene as a set of deformable neural mesh primitives (DNMPs), which are able to capture the complex geometry and appearance of urban scenes.

HRegNet: A Hierarchical Network for Efficient and Accurate Outdoor LiDAR Point Cloud Registration

Fan Lu, Guang Chen^†, Yinlong Liu, Lijun Zhang, Sanqing Qu, Shu Liu, Rongqi Gu, Changjun Jiang

IEEE Transactions on Pattern Analysis and Machine Intelligence (T-PAMI), 2023

Extended journal version of HRegNet

Sparse-to-Dense Matching Network for Large-scale LiDAR Point Cloud Registration

Fan Lu, Guang Chen^†, Yinlong Liu, Yibing Zhan, Zhijun Li, Dacheng Tao, Changjun Jiang

IEEE Transactions on Pattern Analysis and Machine Intelligence (T-PAMI), 2023

This paper proposes a Sparse-to-Dense Matching Network (SDMNet), which performs registration in two stages: sparse matching stage and local-dense matching stage. We design a novel neighborhood matching module to incorporate local neighborhood consensus and the local-dense matching stage is followed for fine-grained performance.

HRegNet: A Hierarchical Network for Large-scale Outdoor LiDAR Point Cloud Registration

Fan Lu, Guang Chen^†, Yinlong Liu, Lijun Zhang, Sanqing Qu, Shu Liu, Rongqi Gu

International Conference on Computer Vision (ICCV), 2021

This paper proposes a hierarchical network named HRegNet, which leverages rich features in deeper layer and precise location information of keypoints in shallower layers for robust and accurate LiDAR point cloud registration. Bilateral consensus and neighborhood consensus are utilized to improve the robustness of the keypoints matching.

PointINet: Point Cloud Frame Interpolation Network

Fan Lu, Guang Chen^†, Sanqing Qu, Zhijun Li, Yinlong Liu, Alois Knoll

Thirty-Fifth AAAI Conference on Artificial Intelligence (AAAI), 2021

We propose a novel framework, namely Point Cloud Frame Interpolation Network (PointINet). Based on the proposed method, the low frame rate point cloud streams can be upsampled to higher frame rates.

MoNet: Motion-based Point Cloud Prediction Network

Fan Lu, Guang Chen^†, Yinlong Liu, Zhijun Li, Sanqing Qu, Tianpei Zou

IEEE Transactions on Intelligent Transportation Systems (T-ITS), 2021

RSKDD-Net: Random Sample-based Keypoint Detector and Descriptor

Fan Lu, Guang Chen^†, Yinlong Liu, Zhongnan Qu, Alois Knoll

Advances in Neural Information Processing Systems (NeurIPS), 2020

This paper proposes Random Sample-based Keypoint Detector and Descriptor Network (RSKDD-Net) for large scale point cloud registration. The key idea is using random sampling to efficiently select candidate points and using a learning-based method to jointly generate keypoints and descriptors.