🀁 Nan Wang   

I am an Algorithm Engineer at the Beijing Academy of Artificial Intelligence (BAAI), where I develop and implement cutting-edge solutions in neural rendering and world models, exploring their potential applications in VR, robotics, and autonomous driving. I earned my Master's in Computer Science from the Tongji University.

profile photo

News

Research

Representative papers are highlighted.

Unifying Appearance Codes and Bilateral Grids for Driving Scene Gaussian Splatting
Nan Wang, Yuantao Chen, Lixing Xiao, Weiqing Xiao, Bohan Li, Zhaoxi Chen, Chongjie Ye, Shaocong Xu, Saining Zhang, Ziyang Yan, Pierre Merriaux, Lei Lei, Tianfan Xue, Hao Zhao
NeurIPS, 2025

We use Multi-Scale Bilateral Grids that improves geometric accuracy in driving scene reconstruction through enhancing photometric consistency.

HOLO: Holistic Lightweight Optimization for Scene Understanding with Auto-Annotation and Multimodal Learning
Xiaoyun Hu*, Xiaohan Yan*, Nan Wang, Xiaowei Song, Gang Wei, Zhicheng Wang
WACV, 2026

We propose HOLO, which includes a large-scale scene description dataset and a lightweight 3D-LLM.

One View, Many Worlds: Single-Image to 3D Object Meets Generative Domain Randomization for One-Shot 6D Pose Estimation
Zheng Geng, Nan Wang, Shaocong Xu, Chongjie Ye, Bohan Li, Zhaoxi Chen, Sida Peng, Hao Zhao
CoRL, 2025, Oral presentation

We use 3D Generative Model to generate the 3D object and then utilize it to estimate accurate 6D pose.

Simulate Any Radar: Attribute-Controllable Radar Simulation via Waveform Parameter Embedding
Weiqing Xiao*, Hao Huang*, Chonghao Zhong*, Yujie Lin, Nan Wang, Xiaoxue Chen, Zhaoxi Chen, Saining Zhang, Shuocheng Yang, Pierre Merriaux, Lei Lei, Hao Zhao
arXiv, 2025

We present SA-Radar, a radar simulation approach that enables controllable and efficient generation of radar cubes conditioned on customizable radar attributes.

ORV: 4D Occupancy-centric Robot Video Generation
Xiuyu Yang*, Bohan Li*, Shaocong Xu, Nan Wang, Chongjie Ye, Zhaoxi Chen, Minghan Qin, Yikang Ding, Xin Jin, Hang Zhao, Hao Zhao
arXiv, 2025

We propose ORV, an Occupancy-centric Robot Video generation framework.

PUGS: Zero-shot Physical Understanding with Gaussian Splatting
Yinghao Shuai, Ran Yu, Yuantao Chen, Zijian Jiang, Xiaowei Song, Nan Wang, Jv Zheng, Jianzhu Ma, Meng Yang, Zhicheng Wang, Wenbo Ding, Hao Zhao
ICRA, 2025

We reconstruct 3D objects using the Gaussian splatting representation and predicts various physical properties in a zero-shot manner.

RE0: Recognize Everything with 3D Zero-shot Open-Vocabulary Instance Segmentation
Xiaohan Yan*, Zijian Jiang*, Yinghao Shuai*, Nan Wang, Xiaowei Song, Wenbo Ji, Ge Wu, Jinyu He, Gang Wei, Zhicheng Wang
ICRA, 2025

We leverage the 3D geometry information and the semantic features to address the challenge of 3D instance segmentation.

Semantic-Guided Gaussian Splatting with Deferred Rendering
Nan Wang, Xiaohan Yan, Xiaowei Song, Zhicheng Wang
ICASSP, 2025

We use semantic features derived from 2D foundation model to revolutionize the material property optimization for 3DGS.

GreedyAgent: A Simple yet Efficient Approach for Meta learning from Learning Curves
Jinyu He, Xiaowei Song, Xiaohan Yan, Nan Wang, Yuqi Miao, Zijian Jiang, Fei Chao, Yan Zhang, Shengchuan Zhang, Rongrong Ji
ICIC, 2024, Oral presentation

A key sub-problem: meta-learning from learning curves is an mature but gradually attention area within the field of meta-learning.

AttenPoint: Exploring Point Cloud Segmentation through Attention-Based Modules
Xiaohan Yan, Nan Wang, Xiaowei Song, Gang Wei, Zhicheng Wang
PRCV, 2024

We combine local and global features of the structures and performance to perform few-shot point cloud semantic segmentation.

Projects

A Real2Sim Pipeline for robotics simulation
Xiaomi
2024-06

We use 3D Gaussain Splating to build a novel robotics simulator, the physical simulation is implemented by ISAAC SIM while the photorealistic render is implemented by 3DGS.

3DGS implementation for Structured-NeRF: Hierarchical Scene Graph with Neural Representation
DISCOVER Robotics
2023-11

I use 3D Gaussian Splatting and nerfstudio to build an implementation, which can accomplish the 3D scene decomposition and Reconstruction.

Pytorch-Lightning Implementation of RandLA-Net && SQN
2023-10

RandLA-Net (CVPR2020) is a light net which can process efficient semantic segmentation for 3D point clouds, and SQN (ECCV2022) achieves promising performance on seven large-scale open datasets under weak supervision schemes.

LLM Science Exam - Use LLMs to answer difficult science questions
Nan Wang, Xiaohan Yan, Xiaowei Song
Kaggle, 2023-10

LLM4Science 🥈 silver medal solution. We make it into a RAG (Retrieval Augmented Generation) task and finetune three Deberta models with different finetuning and combine their output features to infer the right answer.

Short Bio

I am a M.Sc student in Computer Science at CAD Research Center, Tongji University.

show more

Misc

Music 🎶
  • Piano 🎹 (Grade 10 certification from Shanghai Conservatory of Music)
  • Sports 🏃‍♂️
  • Basketball🏀
  • Badminton🏸
  • Swimming 🏊‍
  • Flying Disc🥏
  • Language 💬
  • 中文 🇨🇳
  • English 🇺🇸
  • 日本語 🇯🇵 (learning)
  • Français 🇫🇷 (learning)
  • Others 💡
  • Guangfulin