Deep Q Networks (DQNs) have emerged as a powerful reinforcement learning framework that integrates deep learning with classical Q-learning to efficiently handle high-dimensional decision-making environments. DQNs were initially used in game-playing applications, but today they are showing transformational impact in robotics and real-world control systems. This chapter explains in detail the foundational principles of DQNs such as Q-value approximation, experience replay, and target networks, which make learning more stable and efficient. Practical applications such as Atari games prove that DQNs can learn optimal policies even from raw sensory input. The use of DQNs in robotics is growing rapidly for tasks such as navigation, obstacle avoidance, and precision control. However, the strengths of DQNs such as generalization and autonomous learning are accompanied by some limitations such as high computational cost and training instability. Variants such as Double DQN, Dueling DQN, and Prioritized Experience Replay have been developed to address these issues. The conclusion of the chapter focuses on future perspectives where the integration of hybrid architectures—such as meta-learning, symbolic reasoning, and attention mechanisms—makes DQNs even more robust and intelligent.