Reconfigurable Intelligent Surfaces meet Multi-Agent Reinforcement Learning
The convergence of sixth-generation (6G) wireless networks, vehicle-to-everything (V2X) communications, and Reconfigurable Intelligent Surfaces (RIS) presents unprecedented opportunities for revolutionizing mobility management in high-speed vehicular scenarios. This paper proposes a novel decentralized AI Agent framework that synergistically integrates Multi-Agent Reinforcement Learning (MARL) with predictive Transformer-based agents for dynamic optimization of RIS phase shifts and handover protocols in ultra-reliable low-latency communication (URLLC) scenarios.
Our framework comprises three specialized agents: a RIS Optimization Agent that dynamically adjusts phase shifts for coverage enhancement, a Handover Management Agent that makes proactive handover decisions based on trajectory prediction, and a Resource Allocation Agent that optimizes spectrum and power allocation. We formulate the joint optimization problem as a cooperative Markov Decision Process and employ QMIX and Multi-Agent Proximal Policy Optimization (MAPPO) algorithms for distributed decision-making.
Dynamically adjusts phase shifts for coverage enhancement using reinforcement learning.
Makes proactive handover decisions based on Transformer-based trajectory prediction.
Optimizes spectrum and power allocation through MAPPO distributed decision-making.
Authors: AlHussein A. AlSahati, Dr. Houda Chihi
Research Lab: InnovCOM Lab of Sup'COM Tunisia