Buildathon 4.0 – 2nd Place | ITU FG-AINN AI Native Networks
Team NVK (Charan Velavan & Yogender Raju, SSN College of Engineering) secured 2nd Place at Buildathon 4.0: AI Native Networks & Applications (ITU × FreeStream). The project is an AI-native converged 6G–ATSC 3.0 network architecture for scalable UAV swarm coordination in smart cities. An edge-deployed AI Mission Coordination Server on Raspberry Pi 5 dynamically classifies UAV telemetry and selects the optimal delivery mode — 6G unicast for latency-critical control, 5G eMBMS multicast for group missions, and ATSC 3.0 broadcast for mass distribution — achieving up to 98% bandwidth savings. Submitted as ITU Focus Group document FG-AINN-I-215 and mentored by Dr. P. Kaythry.
Context
ITU Buildathon 4.0 focused on AI-native networks and applications within the Framework of ITU's Focus Group on AI Native for Telecommunications Networks (FG-AINN). The challenge demanded working Proof-of-Concept prototypes integrating real AI and networking for smart city and next-generation communication use cases, judged against international standards bodies.
The Challenge
UAV swarms deployed in smart cities for surveillance, disaster response, and traffic monitoring require both low-latency bidirectional control and efficient distribution of shared mission data to hundreds of drones simultaneously. Unicast-only cellular approaches create backhaul overload; pure broadcast lacks bidirectional control. The challenge was to combine these heterogeneous technologies intelligently at the network edge.
Our Solution
Designed and implemented an AI-native converged 6G–ATSC 3.0 architecture. An edge-deployed AI Mission Coordination Server on Raspberry Pi 5 runs a trained ML classification model that consumes real-time telemetry features (swarm size, spatial density, congestion, priority, latency needs) to dynamically select between 6G unicast (for latency-critical control), 5G eMBMS multicast (for group targeting), and ATSC 3.0 broadcast (for mass mission dissemination). A Python-based UAV simulator models 150–300 UAVs with geo-coordinates, battery dynamics, and congestion. A Streamlit Mission Control Dashboard visualizes AI decisions, bandwidth savings, 2D/3D UAV maps, and predictive analysis in real time.
My Role
Co-architect of the system with Yogender Raju (Team NVK), guided by Dr. P. Kaythry (SSN). Designed the AI architecture and training pipeline, built the UAV Swarm Digital Simulator in Python, deployed the ML model on Raspberry Pi 5, implemented the inter-device communication protocol, and developed the Streamlit Mission Control Dashboard.
Impact & Results
Achieved up to 98% bandwidth savings (~147 MB per operation) vs. unicast-only baseline. Maintained sub-50ms latency for critical 6G unicast traffic while running concurrent ATSC 3.0 broadcast. Successfully coordinated 150–300 UAVs in real-time with sub-100ms AI edge inference. Submitted as official ITU FG-AINN Input Document (FG-AINN-I-215) for the 2025–2028 study period.
Recognition
2nd Place at Buildathon 4.0: AI Native Networks & Applications, organized by the International Telecommunication Union (ITU) and FreeStream Technologies. Official Input Document submitted: FG-AINN-I-215. Collaboration with Edge Beam Wireless. Internal Guide: Dr. P. Kaythry, SSN College of Engineering.