• A Rajaguhan
• Prabin R
• Nivedha S

In this project, we showcase a deep learning model leveraging reinforcement learning to derive control policies directly from high-dimensional sensory inputs. Our approach employs Convolutional Neural Networks trained through both Deep Q-Networks (DQN) and Double DQN (DDQN). Remarkably, our model processes raw pixel data from the classic Atari Pong game within the Arcade Learning Environment, predicting and optimizing future rewards. The distinctive aspect of our methodology lies in its robustness and generalizability. Notably, we find that DDQN consistently outperforms DQN without requiring any modifications to the network architecture or the learning algorithm. Through extensive experimentation, our model consistently surpasses previously documented methods. Surpassing well-established hard-coded bots and even outstripping human experts across all testing scenarios, our results underscore the potential of deep reinforcement learning in complex decision-making tasks. This achievement sets a new standard for autonomous systems interacting in dynamic environments.

Deep Learning, Reinforcement Learning, Deep Q-Networks (DQN), Double DQN (DDQN), Convolutional Neural Network, Future Rewards, Atari Pong, Arcade Learning Environment.

"Atari Pong with DQN and DDQN", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.11, Issue 5, page no.f295-f302, May-2024, Available :http://www.jetir.org/papers/JETIR2405532.pdf

"Atari Pong with DQN and DDQN", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.11, Issue 5, page no. ppf295-f302, May-2024, Available at : http://www.jetir.org/papers/JETIR2405532.pdf