Poker Player

Integrating hardware with software to create a minimal-input AI that can read its own cards and make its own decisions

Inspiration

We like building cool AI's for games, but we wanted to make it even easier for a human to play with a bot.

What it does

Its a Poker AI that plays Texas Hold'em Poker. It can detect its own cards as well as the draw to determine what move to make each turn.

How I built it

We used a Raspberry Pi camera to take pictures of the game. The pictures are fed into an ML-model built using Google Cloud's Auto-ML to translate into computable data. This data is then fed into the Poker's AI algorithm which then determines what move to make.

Challenges I ran into

Our initial goal was to use facial cues of opponents as an input to the Poker's AI so it could factor in the human element of Poker into it's decision making. However, Google's Vision API was not powerful enough to detect subtle changes in facial appearances that occur in a game of Poker.

Accomplishments that I'm proud of

Image recognition of individual playing cards! Google's Auto-ML was incapable of this so we had to collect a lot of data to train our own model to accomplish this.

What I learned

How to use a Raspberry Pi, specifically its camera, to do cool stuff.

What's next for Poker Player

Improving the AI. We focused more on the human element of the project so we could certainly improve the actual AI. We also want to integrate speech-to-text using Google Cloud so the AI can keep track of the game progression by listening to the other players' moves.