On using deep reinforcement learning to reduce uplink latency for uRLLC services