Salvatore Esposito
I'm a researcher and engineer working on autonomous robotic navigation of agents in real-world environments, 3D Vision, and Generative AI. I earned my PhD in Geometric Deep Learning from the University of Edinburgh's CDT in Biomedical AI, where I worked under the supervision of Arno Onken, Oisin Mac Aodha, and Changjian Li. I've contributed to Microsoft's HoloLens and Teams Avatars initiatives, translating complex technical concepts into user-centred AR/VR experiences. My strength lies in bridging research innovation with practical applications, effectively collaborating across engineering, design, and business teams. I combine academic rigour with hands-on development to align technical possibilities with product vision, while fostering collaborative environments that accelerate R&D and drive strategic growth.
Interests
- Robotics and Navigation
- Systems Integration & Roadmapping
- Generative AI & Computer Graphics
- Data Engineering & Feature Pipelines
Education
-
PhD in Geometric Deep Learning (Biomedical AI CDT) University of Edinburgh
-
MRes in Artificial Intelligence University of Edinburgh
Professional Experience
- Developing novel navigation algorithms for robotic systems for minimally invasive procedures to enhance surgical precision and safety
- Leading interdisciplinary research team in creating AI-powered navigation solutions for autonomous surgical interventions
- Led the enhancement of the HMR 2.0 multimodal transformer pipeline for avatars
- Improved motion & audio prediction for more realistic avatar interactions
- Built GAN- and NeRF-based generative models with enhanced surface fidelity
- Proposed a neural rendering architecture for high-quality facial mesh extraction
- Time-series forecasting for CPU utilization in virtual machines
- Designed predictive resource-management solutions
- Integrated NeRFs with caching for improved rendering quality
- Applied ML/AI & statistical methods for financial modelling
- Optimized HPC & cloud pipelines for large-scale data analysis
Publications
VesselSDF: Distance Field Priors for Vascular Network Reconstruction
We propose VesselSDF, a novel approach for reconstructing a 3D signed-distance field from 2D cross-sections…
GeoGen: Geometry-Aware Generative Modeling via Signed Distance Functions
We introduce a new generative approach for synthesizing 3D geometry and images from single-view collections…