Resource Allocation in V2X Networks: From Classical Optimization to Machine Learning-Based Solutions

As one of the promising intelligent transportation frameworks, vehicular platooning has the potential to bring about sustainable and efficient mobility solutions. One of the challenges in the development of platooning is maintaining the string stability, which ensures that there is no amplification of the signal of interest along the platoon chain. String stability is dependent on reliable inter-vehicle communications and proper controller design. Therefore, in this paper, we formulate radio resource management (RRM) problem with the purpose of satisfying the reliability of the vehicle-to-vehicle (V2V) links and string stability of the platoon. We tackle the optimization problem from different angles. First, we devise centralized classical approaches based on difference of two convex functions (d.c.) programming, in which we assume the base station (BS) has full knowledge over the V2V channel gains. In the second strategy, we develop decentralized resource allocation approaches based on multi-agent reinforcement learning (MARL). In essence, we model each transmitter vehicle in the platoon as an autonomous agent that tries to find an optimal policy according to its local estimated information to maximize the total expected reward. We also investigate whether the integration of federated learning (FL) with decentralized MARL algorithms can bring any potential benefits. This comparison between classical and machine learning (ML)-based RRM strategies helps us make crucial observations in terms of robustness, sensitivity, and efficacy of the policies that are learned by reinforcement learning (RL)-based resource allocation algorithms.