UNO RL Arena: JAX-Accelerated Agent

A high-performance reinforcement learning agent for UNO, featuring a highly scalable JAX-based training pipeline and a low-latency FastAPI inference backend.

Comment

Inspiration

The game of UNO is really fun for families. It is also very interesting for Artificial Intelligence. This is because you cannot see what cards the other players have and you have to make decisions with the cards you have. It is a challenge for computers to play this game well. This is why we made a computer program that can play UNO. It uses a kind of memory that helps it remember what happened before so it can make decisions.

What it does

We made a computer program that can play UNO with humans or other computers. It is very good at playing the game. Can make decisions quickly. When a human plays a card the computer program calculates what to do and tells the human what it will do. It does all of this fast so the human does not have to wait.

How we built it

We divided the project into two parts: the computer program that plays the game and the website that people use to play the game. The computer program uses math to make decisions and it can play many games at the same time. We also made a website that people can use to play the game. It talks to the computer program to make decisions.

Challenges we ran into

One problem we had was that the computer program took a time to start up. This was because it had to do some calculations before it could start playing. We fixed this by making the program do some practice calculations before it started playing so it would be ready to go. We also had to figure out how to save the computer programs memory so it could remember what happened before. We did this by using a kind of storage called Redis.

Accomplishments that we're proud of

We are very happy with how fast the computer program can play the game. It can play games at the same time and it can make decisions very quickly. We also made a system that can predict what cards the other players have based on what they have done before.. We made a website that can handle many players at the same time without any problems.

What we learned

We learned a lot about how to make computer programs that can play games like UNO. We learned about math that can help the program make decisions and we learned about how to make the program remember what happened before. We also learned about how to make websites that can handle players at the same time.

Whats next, for UNO RL Arena: JAX-Accelerated Agent

We are going to add some features to the website like the ability to track player statistics and ratings. We are also going to add a system that allows players to log in and play with their friends.. We are going to make it possible for humans to play against each other with the computer program playing as a special kind of player.

Built With

Share this project:

Updates