Note: the demo has been tested in the Firefox and Chrome browsers, Internet Explorer support requires the Chrome Frame plugin.
This demo is a simplified version of the maze domain investigated in several novelty search publications. For simplicity, the search space and the behavior space in the demo are the same: x,y coordinates in a two-dimensional space. That is, each individual is a position (i.e. not a neural network).
Below is a side-by-side comparison of objective-based search (left) and novelty search (right). The blue circles (which will appear as a group after you press "go" below) represent a population of points in the two-dimensional search space. The objective is to evolve a point that reaches the green circle (goal). Mutations are random perturbations of the x and y coordinates of the blue circles; any mutation that causes a point to enter a black square (i.e. wall) is rejected. In this way, walls are constraints in the search space. The "toggle speciation" button causes genotype-based speciation, similar to what happens within NEAT, so that you can see how novelty search and fitness-based search interact with speciation.
The left search process rewards closer distance to the objective, while the right rewards novelty. Everything else is the same.
Return to novelty search users page