Biobloc

Help :: Evolution

Evolution is certainly the most interesting part of the program, your bioBloc will learn itself how to move, according to the form that you gave it. This training is based on genetic algorithms.

1-Move choice

Before anything click on the type of desired move. The numbers on left constitute a cost in evolk, resource which will be used in a future version. Here turn left is selected, whereas walking is already computed.

2-Basics move

These two icons represent selected move (the large one), and basis move (the small one). Here turn left is selected with the 'move choice'. By clicking on the small icon you can choose a walk still calculated or nothing. Indeed it will be more efficient to learn how to turn left on the basis of walking.

3-Simulation's parameters

Here the time taken for the training is decides, based on genetics algorithms. The principle is simple, with a population of 10 and 15 generations, 10 moves are randomly generated, then evaluated according to desired move. Then a new generation of 10 moves is generated by modifying a bit the best displacement previously found. In our example there will be 150 moves tested, it takes from 0.1s to 1s of computing time by moves with a 1.4 java version.

The step duration corresponds to the time takes by your bioBloc to carry out a move's cycle.

4-Start training

Compute start by clicking on the play button.
At any moment, during training, you can pause computing and appreciate the best move found.
This button stops calculation and the best move found is then preserved.

5-Following training

Histogram show the best quality found by generation. The orange bar represents the move which has been just calculated. Thus in our example there will be 10 passages over each bar and they will be 15.