With the promotion of 5G commercial deployment, various application scenarios such as smart transportation more have emerged. Many of latency-sensitive vehicle services, necessitate a prompt response from the network. In this paper, we study the scheduling of downlink communication and computational resources within the constraints of bounded latency in a urban vehicle network. Millimeter wave (mmWave) and cellfree technologies are incorporated into the downlink transmission process. Based on the latency constraints and short packet characteristic of low-latency data transmission, we establish a two-level queue and achievable rate model to minimize system power consumption while ensuring the fulfillment of task latency requirements. Further, we model the resource allocation as a Markov decision process based on the two-level queue and propose a Hybrid Action Multi-Agent Reinforcement Learning (HA-MARL) algorithm to efficiently schedule computational resources and transmission power. Simulation results demonstrate our proposed scheme in terms of successful data transmission probability and power consumption outperforms other baseline schemes. Additionally, it effectively accommodates the coexistence scenario of various latency constraints.
Support the authors with ResearchCoin