In tactical networks, traffic should be delivered in a timely manner satisfying the quality of service (QoS) requirements for survivability and mission success. In this paper, we propose a centralized TDMA slot scheduling based on deep reinforcement learning (DRL) to guarantee the QoS requirements by minimizing end-to-end delay. We consider situations in which mission criticality of tactical traffic is dynamically changing. We introduce a DRL actor-critic algorithm to find a TDMA scheduling policy to minimize the weighted end-to-end delay which is a new metric reflecting the mission criticality of tactical traffic. The simulation results verify that the proposed scheduling policy can guarantee QoS requirements in tactical networks.