AI Learns To Beat Google Pacman Game!
In this video, an AI developed machine learning script is used to teach AI how to play and beat the Google Pacman game. Using DeepQ neural networks as a form of deep reinforcement learning, AI PacMan takes screenshots of the game environment, analyzes them in real time, and makes predictions for his next move in the game!
Make sure to subscribe if you enjoyed! I have some fun videos coming soon.
In this method of learning does the model receive any kind of indicative reward value? Interesting project!
I specifically chose to not add any reward logic to the script that way the AI agent could figure almost everything out itself and determine its own rewards (hence why it took so many episodes)
@@GPTLABS Interesting approach, I saw a guy do an interesting episode on coding a machine learning program for Trackmania, the model had access to in-game stats such as speed and distance to the next corner.
Over 10k trails to complete two levels. Destroy huh? I was expecting it to figure out how to generally play and demonstrate playing indefinitely.
Yes, I didn’t implement any reward logic at all so it had to learn from scratch and decide how to play the game which takes thousands of tries.