Enwar 3.0: An Agentic Multi-Modal LLM Orchestrator for Situation-Aware Beamforming, Blockage Prediction, and Handover Management

A new arXiv paper from Ahmad Nazar, Abdulkadir Celik, and team presents Enwar 3.0, an environment-aware reasoning framework that unifies multi-modal sensing, agentic large language models (LLMs), and context-driven model selection for predictive beamforming, blockage detection, and handover management in millimeter-wave (mmWave) vehicular networks. The architecture builds on prior Enwar iterations by adding a classifier-driven sensor health assessment that detects real-time impairments across camera, radar, LiDAR, and GPS inputs with over 99% accuracy, trained via a novel synthetic degradation pipeline. A chain-of-thought primed LLM, refined with human-in-the-loop feedback, orchestrates specialized agents for beam selection, blockage forecasting, and environment perception, dynamically loading sensor-specific models based on real-time context.

Extensive evaluations across 15 sensor combinations show state-of-the-art performance: beam selection accuracy exceeds 88%, blockage F1-scores surpass 98%, and reasoning correctness reaches 87% on complex decision prompts. This work establishes a scalable foundation for LLM-integrated wireless systems that can reason, perceive, and adapt in real-time, directly addressing the challenge of maintaining robust mmWave connectivity in dynamic vehicular environments. The framework's ability to handle sensor degradation and variable link conditions makes it promising for future autonomous driving and 5G/6G networks.