Hands-On Neuroevolution with Python

Fitness function for maze solvers

In every generation of the evolution, each solution individual (maze solver) is evaluated against all objective function candidates. We use the maximum fitness score obtained during the evaluation of a maze solver against each objective function candidate as a fitness score of the solution.

The fitness function of the maze solver is an aggregate of two metrics—the distance from the maze exit (the objective-based score) and the novelty of the solver's final position (the novelty score). These scores are arithmetically combined using a pair of coefficients obtained as an output from the particular individual in the objective function candidate's population.

The following formula gives the combination of these ...