Efficient Q-Learning over Visit Frequency Maps for Multi-agent Exploration of Unknown Environments