Autonomous Testing of Super Mario Using Behavior Models

The article delves into autonomous testing methods utilized in Super Mario Bros., employing a behavior model approach. This is a follow-up to an ongoing series aiming to perfect the autonomous play and clear levels without human intervention. The key focus is on using a mutation-based input generator, which flips bits in input data to create varied scenarios for testing the game's response, revealing edge situations that might go unnoticed via traditional testing.

Here's a code snippet from the methodology:

import mario
import random

def generate_input(starting_byte, flip_probability, input_length):
    input = []
    next_byte = starting_byte
    for _ in range(input_length):
        for j in range(8):
            if random.random() < flip_probability:
                next_byte ^= (1 << j)
        input.append(next_byte)
    return input

This approach is designed to mimic realistic game play, allowing certain keys to remain pressed over multiple frames, akin to how players hold 'move right' while tapping 'jump'. A collection of paths, represented by input sequences, is maintained and selectively replayed to find an optimal course through the game. A simple fitness function favors paths with the highest x-axis position, but due to potential dead-ends, a diverse set of paths with varying scores is explored to ensure comprehensive testing.

This technique is particularly useful for developers involved in game development or those interested in testing automation, offering insights into efficient exploration of complex state spaces.