Agritech robotics, computer vision and navigation. Oscar was principal investigator, secured the funding and managed the programme.
Teaching machines to understand the physical world.
Oscar's research asks how machines can perceive, map and act in the world from ordinary cameras and sensors, and how to make that work outside the lab.
Research themes
Each theme links work done at the University of Surrey with problems Oscar sees in deployed robots.
Bird's-eye-view mapping and geo-localisation
Turning camera images into top-down maps and locating robots against them, from the ICRA 2022 Best Paper to BEV-SLAM, SpaGBOL and PEnG.
Gaussian splatting and neural rendering
Photorealistic 3D scenes from Gaussian splatting and neural fields, used to make realistic robot simulation affordable.
Hyperspectral and multimodal sensing
Self-supervised learning for hyperspectral imagery (HyperKon, HyperGS, SpecBPP, HYDRA) and localisation from Raman spectroscopy.
Embodied AI and human pose
Visual odometry, navigation learned by reinforcement, and 3D hand, body and sign-language pose estimation.
SeDAR dataset
Sequences recorded by a mobile robot moving through CVSSP's offices at different times, with people, moving obstacles and changing layouts. Released with the IJCV paper on human-inspired localisation.
Funded projects
More than £2.3M won and led as principal investigator or project lead.
Mapping, localisation and sensor fusion for autonomous vehicles, with Parkopedia, StreetDrone and the Transport Systems Catapult.
Contrastive learning and mapping from drone and satellite imagery. Oscar was principal investigator.
Researchers supervised
Doctoral and postdoctoral researchers Oscar has supervised or co-supervised.
- Dr Avishkar SahaInstantaneous mapping and graph-based learning
- Dr James RossSLAM and bird's-eye mapping for autonomous maritime vehicles
- Dr Nimet KaygusuzVisual odometry for autonomous vehicle stability
- Dr Celyn WaltersAutonomous calibration and planning for autonomous vehicles
- Dr Xihan BianVisual navigation using reinforcement learning
- Dr Nivedita BijlaniAnomaly detection using graph neural networks
- Dr Maksym IvashechkinBody pose estimation using kinematics and vision
- Dr Daniel AyubaPrecision agriculture using hyperspectral sensors
- Tavis ShoreCross-view geo-localisation using bird's-eye-view estimation
- Christopher ThirgoodLocalisation and mapping using Raman spectrometry
- Georgina Alcolado NuthallNeRF-based mapping for a robot guide dog