Ning Xu

Senior Machine Learning Engineer at NVIDIA.

prof_pic.png

I am a Senior Machine Learning Engineer at NVIDIA. My work centers on one goal: helping physical agents — autonomous vehicles and robots — perceive and understand the 3D world they move through. Over the past six years I have built spatial perception systems across the autonomy stack, from centimeter-accurate localization, to online and city-scale mapping, to neural scene reconstruction.

At NVIDIA, I work on neural reconstruction for autonomous vehicle simulation. Most recently I helped bring Instant NuRec — a feed-forward 3D Gaussian Splatting model that turns a short multi-camera driving log into a fully simulatable 3D world — into NVIDIA’s AV simulation stack. Previously, I developed BEV-transformer online HD mapping models and multi-city-scale 3D SLAM systems at Nuro, and engineered centimeter-accurate localization for production robotaxis at Motional. I received my M.S. in Robotics from the University of Michigan, advised by Prof. Chad Jenkins, and my B.E. from Beihang University.

I am broadly interested in spatial intelligence: 3D/4D reconstruction, generative reconstruction, geometry foundation models, and neural simulation that closes the loop between the physical and digital worlds. I believe grounding robot foundation models in 3D geometry is key to building agents that can truly reason about — and act in — the physical world.

Selected Publications

  1. arXiv
    instant-nurec-gallery.gif
    Instant NuRec: Feed-Forward 3D Gaussian Reconstruction for Driving Scene Simulation
    Jiahui Huang, Jiawei Ren, Michal Tyszkiewicz, Bjoern Haefner, Michael Shelley, Xin Kang, Seung Wook Kim, Ning Xu, Qi Wu, Janick Martinez Esturo, Shengyu Huang, Nick Schneider, Laura Leal-Taixé, Zan Gojcic, and Sanja Fidler
    arXiv preprint arXiv:2607.14203, 2026
    arXiv Bib Code Website
  2. CVPR
    map-change-camera.png
    Exploring Real World Map Change Generalization of Prior-Informed HD Map Prediction Models
    Samuel M. Bateman, Ning Xu, H. Charles Zhao, Yael Ben Shalom, Vince Gong, Greg Long, and Will Maddern
    In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshop on Autonomous Driving, 2024
    arXiv Bib
  3. RA-L
    geofusion-fetch.jpg
    GeoFusion: Geometric Consistency informed Scene Estimation in Dense Clutter
    Zhiqiang Sui, Haonan Chang, Ning Xu, and Odest Chadwicke Jenkins
    IEEE Robotics and Automation Letters, 2020
    arXiv Bib

Selected Projects

NVIDIA neural reconstruction for autonomous vehicles
Neural Reconstruction for Autonomous Vehicles at NVIDIA
Turning real-world driving logs into fully simulatable 3D Gaussian Splatting worlds for closed-loop AV simulation with NVIDIA Omniverse NuRec — from integrating the feed-forward Instant NuRec model into the production reconstruction pipeline to delivering the camera-only NuRec solution.
Video
Nuro unified perception model
Unified Perception Model at Nuro
Prototyped and developed a BEV-transformer model for online HD map construction, integrated within Nuro's unified camera–LiDAR BEV perception framework — accelerating the deployment of multi-task perception models onto the road.
Blog
Nuro scalable online mapping
Scalable Online Mapping at Nuro
Applied research on fusing prior HD map data with real-time sensor streams, significantly improving robustness and safety against real-world environmental and structural map changes — the foundation of our CVPR 2024 paper.
Blog
Nuro 3D city-scale SLAM
3D City-Scale SLAM System at Nuro
Built and maintained a highly reliable, multi-city-scale 3D mapping and SLAM pipeline — analyzing and optimizing scan matching and parallel graph optimization to unlock massive-scale, physics-grounded map building.
Blog