A Hybrid Sensing and Reinforcement Learning Scheduler for Vehicle-to-Vehicle Communications

Vehicle-to-vehicle (V2V) communications performance depends significantly on the approach taken to schedule the radio resources. When the infrastructure is available, so far the best performing V2V scheduling algorithms are based on centralized approach. In case there is no infrastructure, sensing the resources in a distributed manner to determine whether a specific resource is free performs well. We propose a hybrid solution, where a centralized reinforcement learning (RL) algorithm provides a candidate subset of resources, whereas a distributed sensing mechanism, running on each vehicle, makes the final resource selection. We evaluate the performance of the proposed approach in an out-of-coverage setting and show that it outperforms the state-of-the-art algorithms in highly dynamic scenarios by using the best of both worlds: RL agent provides optimized long-term resource allocations, while the distributed sensing handles temporary and unforeseen network conditions that can not be predicted effectively.