https://shikanchen.github.io/2048AI/

The agent is a depth-3 game tree, designed in a structure that the AI player as a max player and game as a chance player (generating a 2-tile at a random spot). The game tree is evaluated in weights by the expectimax algorithm to compute decisions for the agent.

prob = _matrix.length   prob / (i   j   1);

The current developing heuristic value is account for the size of the matrix and the manhattan distance between tiles.