Project
AI-Driven Converged 6G Network for Scalable UAV Swarm Coordination
Team NVK (Charan Velavan & Yogender Raju, SSN College of Engineering) designed and implemented 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 incoming UAV telemetry and selects the optimal delivery mode — 6G unicast for latency-critical control, 5G eMBMS multicast for group targeting, and ATSC 3.0 broadcast for mass mission dissemination — achieving up to 98% bandwidth reduction and sub-100ms latency for critical traffic. The work was submitted as an official Input Document (FG-AINN-I-215) to the ITU Focus Group on AI Native Telecommunications Networks and secured 2nd Place at Buildathon 4.0 (ITU × FreeStream).
View Project👤My Role & Contributions
Co-Author & System Architect (Team NVK)
Key Responsibilities
- •Co-designed the AI-native converged 6G–ATSC 3.0 architecture for UAV swarm communication
- •Architected and deployed the ML-based AI Mission Coordination Server on Raspberry Pi 5
- •Trained the classification model to select optimal delivery mode (unicast / multicast / broadcast) from telemetry features
- •Developed the Python-based UAV Swarm Digital Simulator modelling 150–300 UAVs with realistic mobility, battery dynamics, and congestion patterns
- •Implemented the bidirectional communication protocol between simulation host and Raspberry Pi edge AI server
- •Built the Streamlit Mission Control Dashboard with real-time AI analytics, 2D/3D UAV maps, bandwidth visualization, and manual test panels
- •Prepared and submitted the official ITU Input Document (FG-AINN-I-215) with full PoC description and results
- •Presented and defended the system at the ITU Buildathon 4.0 virtual event (10–12 February 2026)
Technologies Used
Problem
UAV and drone swarms deployed in smart city environments (surveillance, disaster response, traffic monitoring, infrastructure inspection) require real-time low-latency control and simultaneous receipt of shared mission data. Existing unicast-centric cellular networks are fundamentally inefficient at scale — each UAV receiving individual mission updates creates excessive backhaul load, high latency, and limited scalability. Broadcast technologies like ATSC 3.0 solve the one-to-many problem but lack native support for bidirectional control. No existing system intelligently combines these technologies at the network edge.
Approach
- •Proposed an AI-native converged 6G–ATSC 3.0 architecture where an edge-deployed AI Mission Coordination Server dynamically selects delivery modes based on context
- •Deployed a trained ML classification model on Raspberry Pi 5 (edge AI server) that consumes telemetry-derived features: swarm size, payload size, priority, spatial density, congestion level, latency requirements, packet loss estimates, and predicted congestion trends
- •Implemented adaptive confidence thresholding: high-confidence AI decisions auto-apply; a rule-based fallback ensures safe operation under low-confidence or resource-constrained conditions
- •Built a Python-based UAV swarm digital simulator modelling 150–300 UAVs with realistic geo-coordinates, kinematic motion, battery dynamics, spatial clustering, conflict generation, and congestion evolution
- •6G unicast for latency-critical individual commands and telemetry feedback; 5G eMBMS multicast for targeted group missions; ATSC 3.0 broadcast for mass mission data distribution to the entire swarm
- •Developed a real-time Mission Control Dashboard in Streamlit — displaying AI decision analytics, network performance metrics, bandwidth savings visualization, 2D/3D UAV geolocation maps, battery monitoring, and manual test panel with predictive congestion analysis
- •Demonstrated scenario comparison: pure unicast (220–230ms avg latency, 30% congestion at 200 UAVs), non-intelligent multicast (>40% congestion, unstable latency), and the proposed AI-driven mixed strategy (comparable latency, ~147 MB bandwidth savings, ≈96% reduction, 32–33% congestion)
Outcome
- •Achieved up to 98% bandwidth savings without compromising QoS for mission-critical operations
- •Maintained sub-50ms latency for critical 6G unicast traffic while supporting bandwidth-intensive ATSC 3.0 broadcast simultaneously
- •Successfully simulated and managed 150–300 UAVs in real-time with sub-100ms AI inference latency on edge hardware (Raspberry Pi 5)
- •Demonstrated effective coexistence of heterogeneous network technologies (6G, 5G eMBMS, ATSC 3.0) in a single AI-orchestrated architecture
- •Submitted official Input Document FG-AINN-I-215 to the ITU Focus Group on AI Native Networks (FG-AINN) for the study period 2025–2028
- •Secured 2nd Place at Buildathon 4.0: AI Native Networks & Applications (ITU × FreeStream), Virtual, February 2026
- •Validated as a viable foundation for commercial drone delivery, emergency response coordination, agricultural monitoring, and smart city infrastructure
