Detailed description of project

DARIUS is an innovative project introducing an edge-enabled Federated Learning (FL) framework to revolutionize urban planning in European smart cities. Unlike traditional cloud platforms that centralize raw data, DARIUS preserves data sovereignty by training AI models locally within municipal browsers using WebAssembly and WebGPU acceleration. This multi-city collaborative approach allows municipalities to share intelligence—rather than sensitive data—to build robust predictive models.

A core differentiator is the DARIUS “what-if” simulation engine, which enables planners to quantify the impacts of infrastructure changes—such as reduced $CO_2$ emissions, decreased congestion, and improved parking search times—before physical deployment. Integrated into DOTSOFT’s NOON IoT platform, DARIUS shifts the focus from historical dashboards to evidence-based, predictive planning across mobility, air quality, and waste management.

Aligned with the EU Green Deal and AI Act, the project promotes trustworthy, low-latency Edge AI. By bridging the gap between research and market (TRL 4-5), DARIUS provides a scalable solution for over 300 existing deployments, fostering data solidarity and sustainable urban development without the high costs of data centralization.

Type and scope of work provided

• Federated AI Framework: Development of a browser-based FL system using WebAssembly and WebGPU to enable multi-city collaborative training without transferring raw data.
• “What-If” Simulation Engine: Creation of a predictive tool to quantify urban planning outcomes—such as $CO_2$ reduction, congestion levels, and parking availability—prior to physical investment.
• Cross-Domain Orchestration: Integrating datasets from mobility, air quality, waste, and energy into DOTSOFT’s NOON IoT platform.
• Technical Benchmarking: Utilizing the dAIEDGE Virtual Lab to measure energy efficiency, latency, and model accuracy on real edge devices (e.g., Jetson Orin).
• Privacy & Compliance: Ensuring data sovereignty and alignment with the EU AI Act through on-device computation and low-latency inference.
• Pilot Validation: Demonstrating the solution in live municipal environments to ensure scalability and market readiness.