OptANN

Description

OptANN is a separated Khepera simulator and a simple GA (Genetic Algorithm) connected to an more advanced ANN (Artificial Neural Net) implementation. The simulator can handle multiple Khepera robots and a environment that changes during simulation. It simulates round obstacles, walls and light bulbs using a technique with pre-recorded sensor values sampled with a Khepera robot for different objects, angles and distances. The ANN uses a idea where net nodes has an activation priority instead of belonging to a specific layer, which opens up and simplifies more complex ANN arcitectures.

User interface

The simulation environment can be runned with and without graphical interface in order to optimize for speed or evaluation, you can even exclude the GUI code during compile time for greater performance.

With the graphical interface which is using GTK (GIMP Tool Kit) and libggi (General Graphic Interface) you have the possibility to monitor:

• the activation values of the running neural net.

• the robots interaction with the environment with functions as pause, step, graphical trace with different plotting styles.

• the sensor values that are generated both as plots and as colorcoded sensoractivation mapped to a robot figure (where the color represent the activation of the specific sensor).

You can move the robot in the environment and take snapshots for later studies.