This project was my entry to a kind of science fair hosted by Testo in 2015 The entry consisted of a "uArm" robot arm that would autonomously play a game of "Breakthrough" against a human opponent. The board state was detected using a webcam. Competative Gameplay was enabled through implemention of Monte-Carlo simulation with a heuristic function being used after a certain depth of the game tree. Throughout the developement process, both ad-hoc and learned heuristic functions where created. In one instance, a neural network was evolved (through NEAT-python) using selfplay, which was used to evaluate game states.
For the final presentation however, one of the ad-hoc heuristic functions was deemed best. The final agent could trade peaces quite well, however it wasn't as good at actually winning games.

The entry was finally awarded the 2nd place prize in the expert category. The first place was a digital tic-tac-toe console