Mobility Aware Channel Allocation for 5G Vehicular Networks using Multi-Agent Reinforcement Learning

Reinforcement learning is a machine learning technique that focuses on exploring an uncharted territory exploiting of current knowledge. This paper proposes a Mobility Aware Channel Allocation (MACA) algorithm for 5G Vehicular Networks using a combination of Multi-Agent Reinforcement Learning (MARL) and Semi-Markov Decision Process (SMDP). In this work, we use multiple autonomous agents operating in a common environment to address the sequential decision-making problem to optimize the long-term rewards. In MACA, first we predict the mobility of vehicles using Teammate-Learning model as it allows the vehicles to cooperate and collaborate with each other without prior coordination. Secondly, during SMDP resource allocation phase, MARL inputs are applied to the Action Selection model for each vehicle based on their priorities. This is done at Road-Side Units (RSUs). Through numerical results and evaluations, we verify that the proposed method demonstrates efficient channel allocation and high packet delivery ratio as compared in the scenario of vehicles with multiple (high, medium, and low) priorities to existing conventional SMDP and Greedy algorithms.