Reinforcement learning-based distributed channel access for delay optimization