In this paper, we propose a full hardware-based deep Q-networks (DQNs) model for an Autonomous Agents System. DQNs are based on deep neural networks. They are particularly advantageous in complex and high-dimensional environments with large state spaces and intricate decision-making processes. When the DQNs are trained, they can generalize their learned policies for unseen situations or variations of the environment. DQNs are potentially more versatile in handling diverse scenarios without requiring manual adjustments. To learn the variations of the environment, we utilize a target network and an experience replay buffer. Besides, we use the less intensive computation SGD algorithm and apply advanced techniques, such as LeakyReLU, Experience Replay with the last state, and Changing Learning Rate after interactions with the environment to ensure the stability of the model and improve the hardware resources and processing latency. Evaluation results using OpenAI Gym demonstrate that the proposed algorithm and its FPGA implementation complete CartPole-v0 with a time of 0.103 (s) using a neural network with two hidden layers and the number of hidden-layer nodes is 8. This design is synthesized on the Virtex-7 VC707 Evaluation Board (xc7vx485tffg1927-1) with a gain frequency of 250 (MHz) and a power consumption of 0.944 (W).