Salvatore Esposito
Researcher & Engineer
University of Edinburgh
I'm a researcher and engineer working on 3D reconstruction, generative AI including world models, and SLAM. I built the first monocular SLAM pipeline for bronchoscopy, which maps the lung and tracks the camera inside it from monocular video. The same 3D reconstruction and generative AI work took me through two years at the Microsoft Mixed Reality Lab, building for HoloLens and Microsoft Teams, and my systems infrastructure work at Huawei and quantitative research at American Express added the habit of making things fast.
Interests
Education
-
PhD in Geometric Deep Learning (Biomedical AI CDT) University of Edinburgh
-
MRes in Artificial Intelligence University of Edinburgh
Professional Experience
Postdoctoral Research Fellow
University of Edinburgh
2025 – Present
Edinburgh
Postdoctoral Research Fellow in Autonomous Robotics Surgery
- Built the first monocular SLAM pipeline for bronchoscopy, so a surgical robot can map the airways and localise itself inside the lung from a single camera
- Built ROOM, the physics-based simulator that generates the photorealistic training data this work depends on, and published it at ICRA 2026
- Research covers 3D reconstruction, generative AI, and SLAM
Research & Development
Microsoft
2022 – 2024
Zurich & Cambridge
Visiting Researcher, Zurich (2022–2024)
- Improved the avatars in HoloLens and Microsoft Teams, leading the rework of the HMR 2.0 multimodal transformer pipeline
- Made avatars move and sound more like the people driving them, by sharpening motion and audio prediction
Research Intern, Cambridge (2022)
- Produced 3D meshes clean enough to use downstream, with GAN- and NeRF-based generative models; Microsoft released the work as GeoGen
- Turned noisy renders into high-quality facial meshes, with a neural rendering architecture I designed
Research Intern
Huawei
2023
Edinburgh
Systems Infrastructure Research
- Made it possible to size virtual machine fleets to real demand, forecasting CPU utilization with time-series models
- Designed resource management that allocates capacity ahead of demand
Research Intern, Cambridge (2022)
- Improved NeRF rendering quality by integrating it with caching
Quantitative Researcher
American Express
2022
London
Quantitative Analysis
- Built the financial models behind large-scale analysis, using machine learning and statistical methods
- Made that analysis run faster, by optimizing its HPC and cloud pipelines
Publications
* Equal contribution
arXiv 2026
Dense Endoluminal Mapping and Localisation from Monocular Bronchoscopy
arXiv preprint arXiv:2610.16570, 2026
MICCAI 2025
VesselSDF: Distance Field Priors for Vascular Network Reconstruction
International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), 2025
Awards
CDT Doctoral Scholarship in Biomedical AI
UKRI
2020
Contact Me
Working on something in robotics, 3D vision, or generative AI? Get in touch.