Junsheng FuPh.D.

Autonomous Driving & Physical AI

Technical Expert at ZenseactGothenburg, Sweden

I lead research initiatives at Zenseact to build long-term competence in end-to-end driving. My work connects geometric and learning-based approaches to localization, online mapping, and scene understanding for autonomous vehicles. Previously, I worked in computer vision at Nokia Research in Finland and hold a Ph.D. in Computer Vision.

Curriculum vitae Google Scholar GitHub LinkedIn
Portrait of Junsheng Fu

News

  • Opening keynote at the ECCV WMEAI Workshop: ‘What Should a World Model Represent in Autonomous Driving?’

  • Beyond Chamfer Distance: Granular Order-Aware Evaluation Metric for Online Mapping — NeurIPS 2026.

  • OpenCVL: An Open, Diverse, and Large-Scale Dataset for Fine-Grained Cross-View Localization — ECCV 2026.

  • QueryOcc: Query-Based Self-Supervision for 3D Semantic Occupancy — CVPR 2026.

  • Invited talk at TU Delft’s Intelligent Vehicles Lab: ‘Engineering Physical AI for Autonomous Driving.’

All newsLess news

  • B2F-Map: Crowd-Sourced Mapping with Bayesian B-Spline Fusion — ICRA 2026.

  • Began co-supervising the Safe-VLA industrial Ph.D. project at Chalmers, connecting linguistic reasoning with safe trajectory planning for autonomous driving.

  • Began co-supervising a WASP industrial postdoctoral project with Linköping University on fusing local crowdsourced maps into a global HD map.

  • Localization Is All You Evaluate: Data Leakage in Online Mapping Datasets and How to Fix It — CVPR 2024.

Research

Vision Language Action Model & World Model

Learning-based driving systems, world representations, and the connection between linguistic reasoning and safe trajectory planning.

Online Mapping & Localization

Building and evaluating maps, fusing crowdsourced observations, and localizing vehicles in complex road environments.

3D Scene Understanding & Representation Learning

Geometric and semantic representations, multimodal learning, and self-supervision for understanding the driving scene.

Publications

All publications
NeurIPS

Beyond Chamfer Distance: Granular Order-Aware Evaluation Metric for Online Mapping

C. B. Lehocine, A. Lilja, J. Fu, and L. Hammarstrand

Order-aware evaluation of online mapping beyond Chamfer distance.

CVPR

QueryOcc: Query-Based Self-Supervision for 3D Semantic Occupancy

A. Lilja, J. Lan, J. Fu, and L. Hammarstrand

Query-based self-supervision for learning 3D semantic occupancy.

ECCV

OpenCVL: An Open, Diverse, and Large-Scale Dataset for Fine-Grained Cross-View Localization

Z. Xia, M. Zaffar, J. Fu, A. Alahi, and J. F. P. Kooij

An open dataset for fine-grained localization across viewpoints.

ICRA

B2F-Map: Crowd-Sourced Mapping with Bayesian B-Spline Fusion

Y. Xie, Y. Xia, E. Stenborg, J. Fu, A. Beauvisage, G. E. Garcia, T. Wu, and G. Hendeby

Bayesian B-spline fusion for crowdsourced mapping.

CVPR Workshop

Exploring Semi-Supervised Learning for Online Mapping

A. Lilja, E. Wallin, J. Fu, and L. Hammarstrand

Semi-supervised learning for online mapping.

CVPR Workshop

NeuRadar: Neural Radiance Fields for Automotive Radar Point Clouds

M. Rafidashti, J. Lan, M. Fatemi, J. Fu, L. Hammarstrand, and L. Svensson

Neural radiance fields for automotive radar point clouds.

CVPR

Localization Is All You Evaluate: Data Leakage in Online Mapping Datasets and How to Fix It

A. Lilja, J. Fu, E. Stenborg, and L. Hammarstrand

Examining data leakage and evaluation practices in online mapping.

ICCV

Zenseact Open Dataset: A Large-Scale and Diverse Multimodal Dataset for Autonomous Driving

M. Alibeigi, W. Ljungbergh, A. Tonderski, G. Hess, A. Lilja, C. Lindstrom, D. Motorniuk, J. Fu, J. Widahl, and C. Petersson

A large-scale multimodal dataset supporting autonomous driving research across perception, localization, and mapping.

Patents

Latest 8 grants & applications

All patent records
US patent application

Method for Determining a Current Position of a Vehicle on a Navigation Map

S. M. R. A. Rahuman, S. Jathavedan, A. Beauvisage, J. Fu, and M. Granath

US application 19/324,613

US patent application

Streetlight Blocks Enhanced Adaptive High Beam Control

J. Fu and P. Lofter

US application 18/897,429

US patent application

Road Geometry Estimation for Vehicles

A. Lilja, J. Fu, and L. Hammarstrand

US application 18/816,542

US patent

Hypothesis Inference for Vehicles

J. Fu and A. Beauvisage

US patent 12,227,195

US patent application

Model-Based Road Estimation

O. Eriksson, L. Hammarstrand, Y. Xia, S. Madhusudhana, V. Kovbas, and J. Fu

US application 18/739,846

US patent application

Vehicle Localization

J. Fu and M. Buder

US application 18/621,667

US patent application

Operational Design Domain Management for Vehicles Having Automated Driving Systems

J. Fu, E. Stenborg, and G. G. Jaime

US application 18/619,975

US patent

Vehicle Localisation

J. Fu, H. Zhang, T. Gustafsson, A. Schindler, E. Sesma Caselles, E. Steinmetz, P. Kielén, A. Beauvisage, and J. Lin Sörstedt

US patent 12,085,403

Dates, identifiers, and grant/application categories follow my October 2026 CV.

Research Leadership

Connecting industrial research with academic collaboration and the next generation of researchers.

17Master's theses supervised
3Industrial PhD
2Industrial PostDoc
26Patent grants & applications
2022–2025Zenseact Gold Inventor

Research supervision

I co-supervise industrial Ph.D. students and postdoctoral researchers with Chalmers and Linköping University.

  • Safe-VLALinguistic reasoning and safe trajectory planning · Chalmers
  • Multimodal learning & MapSenseRepresentation learning and online mapping · Chalmers
  • Crowdsourced HD mappingLocal-to-global map fusion · Linköping University

Talks & Teaching

  1. What Should a World Model Represent in Autonomous Driving?

    Opening keynote · ECCV Workshop WMEAI

  2. Engineering Physical AI for Autonomous Driving

    Invited talk · TU Delft, Intelligent Vehicles Lab

  3. Driving the Future: Discovering the Technology Behind Autonomous Cars

    Guest lecture · Chalmers University of Technology

  4. Advances in NeRFs, Road Estimation, and Mapping for Automated Driving Systems

    Invited talk · Polytechnic University of Turin

Open Source

Selected projects from before 2018.

All repositories
Street-view camera pose estimation with a reference point cloud

Localization · Computer vision

Camera Pose Estimation

Estimating 6-DoF camera pose from street-view imagery and LiDAR point clouds.

Road image and semantic segmentation of the drivable surface

Deep learning · Perception

Road Segmentation

Learning to identify drivable road surfaces with fully convolutional networks on KITTI.

Let's connect.

For research discussions, academic collaboration, and invited talks.

junsheng.fu@zenseact.com