Graybx, led by Luigi d’Ovidio, is developing an AI training platform designed to handle complex LiDAR point‑cloud data, an essential technology for robotics, autonomous vehicles and smart‑infrastructure applications. Supported by a deep‑tech grant from Bpifrance, the company set out to enhance its platform with advanced model‑training capabilities, greater explainability and improved usability for both technical and non‑technical users. To accelerate this work, Graybx partnered with the National Edge AI Hub, led by Newcastle University, to access research expertise, testing facilities and industry engagement opportunities.
The challenge facing Graybx was twofold: LiDAR data is computationally demanding, and existing model architectures such as VoxelNet and CenterPoint require significant optimisation before they can be deployed efficiently on edge devices. Additionally, industries relying on LiDAR, such as transport, logistics and manufacturing, need models that are not only accurate but also interpretable and robust. The National Edge AI Hub supported Graybx in tackling these challenges by helping design and validate the platform’s new LiDAR module, offering feedback on key features such as data visualisation and explainability tools.

A core part of the collaboration involved running experimental training pipelines using the KITTI dataset to identify compact, high‑performing model variants. This work aims to demonstrate the potential to reduce model size by at least a factor of five, making advanced 3D perception models far more deployable in real‑world settings. Beyond the technical work, the Hub also helped connect Graybx with industry partners who will pilot the platform on operational LiDAR use cases, providing vital commercial validation and helping raise the platform’s Technology Readiness Level.
Through this partnership, Graybx is positioned to deliver a more powerful, efficient and industry‑ready LiDAR AI solution, supported by the National Edge AI Hub’s mission to bridge cutting‑edge research with practical impact.