The safe integration of Unmanned Aerial Vehicles (UAVs) into U-space requires far more than conventional simulations. It demands high-fidelity Digital Twins capable of accurately modeling wireless propagation physics, AI-based sensing, and autonomous flight dynamics in complex urban environments. Within the framework of the EUSOME project, the High-Performance Computing infrastructure ARIS, operated by GRNET, plays a central role in enabling this advanced research.
The scientific and computational requirements of the EUSOME models are exceptionally demanding. Ultra-massive MIMO propagation modeling involves handling approximately 38 million unknowns, while UAV trajectory optimization requires managing more than 2 million constraints per scenario. Such complexity makes high-performance computing essential, positioning ARIS as a key enabler of large-scale simulations and AI-driven experimentation.
By leveraging ARIS Fat Nodes equipped with 512GB RAM and the A100 GPU partition, researchers are able to scale their applications efficiently across multiple computing nodes. The infrastructure supports iterative numerical solvers, memory-efficient structured representations, and distributed workloads, ensuring stable and high-performance execution of computationally intensive workflows.
The capabilities provided by HPC ARIS go far beyond what standard computing systems can offer. The infrastructure enables terabyte-per-second scale aggregate bandwidth across nodes, stable training of high-resolution multimodal AI models combining vision and radar data up to 1024×1024 resolution, and large-scale Monte Carlo validation of autonomous UAV flight safety. These features significantly accelerate the transition from theoretical modeling to real-world validation.
The data scale of the workflow further highlights the importance of high-performance computing. Input datasets include 3D urban meshes for ray tracing and high-resolution multimodal sensor data, while the ray tracing benchmarking phase alone can generate up to 5 TB of data, with production targets exceeding 50 TB. AI training involves multiple runs of 100,000 steps each, requiring repeated access to large datasets, while flight optimization evaluates up to 100,000 UAV scenarios, each satisfying more than 2 million constraints.
The outcome of this HPC-enabled research is the development of high-fidelity UAV Digital Twins capable of validating sensing systems, communication performance, and flight safety in realistic urban airspaces. Through the computational power of ARIS, researchers can test and optimize autonomous UAV operations with a level of accuracy and scale that was previously unattainable.
By supporting the EUSOME project, GRNET’s HPC ARIS infrastructure demonstrates the strategic importance of national supercomputing facilities in advancing European research and innovation. The platform serves as a critical bridge between advanced theoretical models and real-world autonomous UAV deployment, reinforcing the role of high-performance computing in shaping the future of safe and reliable urban air mobility.
