Scotland Yard is an asymmetric hide-and-search-based board game with imperfect information. It is a two-sided game, searchers, called detectives trying to catch the hider, called Mr. X. This proves to be an imperfect environment since the location of Mr. X is hidden. Previous solutions to play hide and search games all involved using artificial intelligence and graph-search algorithms. This work aimed to explore a different way to approach these games using Deep Q-Learning as a solution. Many different model architectures were considered and all their performances were evaluated. Bots for Mr. X and the detectives was developed and their nuances observed. We believe that Deep Q-Learning could in fact prove to be a good solution in the case of solving such games in the future, especially those with large search-spaces.
The Deep Q learning models are located inside the londonlaw/aiclients/deep_learning folder.
https://github.com/anyc/londonlaw has been adapted for the purpose of our project.