DIREC project

REINS: Adaptive AI for Industry – Without the Cloud

Project impact

Imagine an algorithm that can analyze the soil in a digging bucket or filter out unwanted sounds in your headphones responding to changes in the environment — without ever connecting to the internet.

The REINS project (Runtime Reconfigurable Embedded Intelligence for Resource-Constrained Cyber-Physical Systems) unites B&O, Leica Geosystems, the Alexandra Institute, DTU, and SDU in a shared mission: to develop AI that runs efficiently on devices that operate in dynamic environments with very limited computing power. The result is technology that saves energy, responds instantly, and can operate anywhere.

DIREC’s support for REINS reflects our ambition to act as a catalyst for innovation that extends beyond the laboratory and delivers tangible benefits to industry and society. The project bridges advanced research with real-world applications, positioning Denmark at the forefront of a global shift away from energy- and data-intensive cloud-based solutions.

PROJECT DATA

Project name

REINS – Runtime Reconfigurable Embedded Intelligence for Resource-Constrained Cyber-Physical Systems

Project period

2025-2027

Funding

DKK 4,000,000

Scientific mission

REINS is pioneering a new generation of AI that adapts in real time—without dependence on internet connectivity or massive data centers. Known as runtime reconfigurable embedded AI, this technology brings intelligence to small, local devices with constrained resources. For instance, B&O headphones will be able to fine-tune sound quality in response to changing noise levels in the environment. Likewise, Leica Geosystems’ sensors and robotic arms will be capable of identifying materials and adjusting their actions accordingly.

By combining techniques such as Neural Architecture Search and MAPE-K control loops, REINS not only enables cutting-edge performance but also advances the development of responsible, explainable AI.

Project Participants

	Xenofon Fafoutis – Professor – Technical University of Denmark
	Mikkel Baun Kjærgaard – Professor – University of Southern Denmark
	Luca Pezzarossa – Associate Professor – Technical University of Denmark
	Mahyar Tourchi Moghaddam – Associate Professor – University of Southern Denmark
	Alexandre Alapetite – Principal Software Solutions Architect – Alexandra Institute
	Dan Saatrup Nielsen – Senior AI Specialist – Alexandra Institute
	Frederik Nør Larsen – Team Lead R&D – Leica Geosystems
	Martin Boegeskov Valentinsen – Chief Architect – Leica Geosystems
	Pablo Martinez-Nuevo – Head of AI – Bang & Olufsen
	Sven Ewan Shepstone – Senior Specialist (AI/ML) – Bang & Olufsen

Partners

This project directly aligns with B&O’s innovation focus, advancing the use of TinyML in next-generation headsets, earbuds, and compact portable speakers. By harnessing runtime-configurable edge AI, we enable intelligent, adaptive experiences that personalize sound and functionality for each user — all within the constraints of low-power, resource-limited devices.

Pablo Martinez-Nuevo
Head of AI
B&O

Leica aims to leverage AI to enhance operational efficiency and precision in material detection, volume estimation, weight estimation, and tool recognition within excavation and dozing contexts. The project is of strategic importance to Leica as it aligns with the company’s commitment to technological innovation and customer workflow optimization, positioning Leica at the cutting edge of AI applications in the engineering sector.

Martin Valentinsen
Chief Architect
Leica Geosystems A/S