December 2019
Intermediate to advanced
368 pages
11h 10m
English
As we have mentioned, navigation through the maze is a deceptive problem that requires a different approach to define the fitness function. In Chapter 5, Autonomous Maze Navigation, we presented you with a particular maze configuration that produces areas with strong local optima of goal-oriented fitness scores. As a result, the training process can be trapped inside these areas and will fail to produce a successful solution. The Novelty Search optimization method was devised to address issues with deceptive fitness function landscapes.
Novelty Search rewards the novelty of a solution rather than its proximity to the ultimate goal. Furthermore, the novelty metric, which is used to calculate the fitness score of ...
Read now
Unlock full access