Description
If you've never played 2048 before, take a jab at it here ! The goal is to combine like-blocks until a single block's sum equals or exceeds 2048.
For one of my AI projects, I created an adversarial search algorithm to solve the 2048 game over 90% of the time with less than 0.2 seconds per turn. It utilizes Alpha-Beta pruning for computational optimization as well as Expectiminimax to model the stochastic nature of the game.
For the heuristic functions, I chose random strategies that often work for human players (i.e. aim to minimize the number of blocks, aim to combine higher numbered blocks, try to keep blocks on one side) as well as incorporating some from Kohler et al. (2019) .
To keep the time under 0.2 seconds per turn, I use iterative depthening search in order to maximize the computational ability of the AI within the given time range.