The aim of the Palamedes project is to determine if it is possible to create expert game AIs for popular backgammon variants using techniques similar to what is used by strong backgammon programs. The project currently supports the Greek Tavli variants (Portes, Plakoto, Fevga) as well as other popular variants like Narde, Nackgammon, Hypergammon, Takhteh and Tapa. Each game’s expert AI is build using neural networks trained by millions of self-play games using advanced reinforcement learning techniques.

Normally backgammon games are played in a 24-point board using 15 checkers or men per side and 2 six-sided dice. This project attempts to extend the popular backgammon games to arbitrary configurations. Games can be played using different board sizes, number of checkers, and number of die sides. Smaller board configurations make the game easier for players and AI algorithms, while larger board configurations make it more challenging.