Opportunistic Slicing Through Deep Reinforcement Learning for the Autonomous Tram Service

The use of a radio resource partitioning algorithm to meet mission critical requirements is investigated in this paper. Applications such as autonomous tram have stringent requirements in terms of both reliability and latency. Rather than resort to ultra-reliable low-latency communications, which are still missing on the market, the emerging radio access network (RAN) slicing technology can be exploited to provide mission-critical level service by reserving to the slice the correct amount of resources. This amount is computed by means of a deep reinforcement learning approach, aiming at minimizing the amount of resources to be allocated to the tram slice while still meeting its stringent requirements.