

We found that the Q-learning agent without function approximation averaged a score of 19.8 across 10 games after 48 hours of training (4300 iterations). We compared each model by looking at their average scores (after training), where the score equals the furthest distance traveled by the agent in a game. We evaluated the performance of a minimax agent, a Q-learning agent, and a Q-learning agent with function approximation. Central to the goal of our project is the implementation of several different agents to tackle Crossy Road, a modern mobile game similar to Frogger. In recent years, many AI breakthroughs have been demonstrated through video games. You're welcome to fork this repository over to implement more tweaks and algorithms to play this Crossy Road game! Introduction This can even go for two players who both have achieved identically the highest scores of the day, meaning that there can be more than one grand prize winner.A project made to explore the differences between various artificial intelligent techniques to score as many points as possible with Crossy Road.


The highest rewards give out tokens, which can be used to unlock token-only mascots. Rewards are distributed based on the ranks players reach by the time the event ends. It resets every 24 hours at 0:00 GMT/UTC (7:00 PM EST). Each day, players must play as a pre-determined mascot on a pre-generated level, and must have internet connection when starting and ending a run.
