A simple heuristic search method for the automatic generation of neural-based game artificial intelligence architectures in Ms. Pac-Man

In this work, we develop a game controller called HillClimbingNet (Hill-Climbing Neural Network) for playing Ms. Pac-man that combines the hill-climbing concept and simple feedforward neural network. The computational experiments have been conducted to evaluate and compare the proposed algorithm against Random Direction (RandDir) and Random Neural Network (RandNet) systems. According to the simulation results, HillClimbingNet has achieved an average score of 6290, but only 439 and 735 on the RandDir and RandNet, respectively. HillClimbingNet has a very good performance.