RLAB: A Reinforcement Learning-based Adaptive Broadcasting for Vehicular Ad-hoc Networks

Effective context-aware broadcasting of information to the areas of interest (AoI) is a challenging problem in vehicular ad-hoc networks. It is usually assumed that the information about these AoI are a priori known, either by a centralized source of information or by the entire set of vehicles. In this paper, we propose a self-adaptive broadcast scheme based on distributed reinforcement learning, in which the vehicles are able to collaboratively tune the rate of their broadcast based on the network dynamics and without the initial knowledge about geographical distribution of AoI. The proposed approach enables a more practical implementation of distributed context-aware broadcasting, where no global information and only a partial synchronization is required. The convergence and broadcast performance of the proposed learning system is evaluated using simulations for several setups. These results show a significant improvement, in temrs of number of useful broadcasts and delay, over the existing approaches, such as gossip-based broadcasting.