DQN Atari AI
In this project I created a DQN AI that plays the Atari game Asterix in order to find any bugs or errors in the game. This AI is a Reinforcement Learning AI that learns through trial and error. Throughout its learning it will adjust to collect object that increase high score and avoid objects that cause it to lose the game and points.
Fig: 1. Oates 2021. Example of DQN AI
This was a 5 week university project at Falmouth University, created with Python.
Features
-
DQN AI - the AI system that plays the game.
-
Genetic Algorithm parameter optimisation system - used to find the best fixed parameters that can be used during training.
-
Reward measuring system - this system measures the current reward of agent to tell if it should save or ignore the current training data
-
Extraction system - this system monitors the agent's system as an extra layer of security on if it should save the current training data.
-
Custom data filter - this system filters data from generated csv to ensure that the agent's progress can be plotted properly.
-
Graph plotting system - this system plots graphs to show the progress of the AI training and works with the custom filter.
-
CSV generator - this works with the AI system and will record and store the progress of the AI onto csv files.
Figure List
Figure 1: Oates 2021. Example of DQN AI. [Video]