The following series of Java applets are designed to illustrate different features of the SNMS model. This applets were complete by students in the REU progam (Summer 2005). (Note a Java plug-in is required to view the applets.)
The following applet shows the behavior of a trained model (grey) versus the target behavior (red).
This simulation shows the relationship between target and the trained movement in the arm model with the use of a biologically realistic neural muscular model. Fitness is a measure of how close the actual motion follows the target motion. Over time the genetic algorithm continually improves/trains the model resulting in more and more accurate movements. The genetic algorithm's behavior is very roughly analgous to the process a baby goes through in learning to control its limbs. The genetic algorithm has a target movement it is trying to achieve; through repeated attempts it slowly learns how to create the desired movement. The figure on the right of the diagram shows the relationship between the ideal and actual movements of the human arm graphically. Again, a high fitness means the two lines will be very similar. Several interesting features to note of the trained motion include the “S” shaped pattern at the beginning of the simulation, the bouncg near the end of the simulation, and the extended hold at the peak of the movement. Some of these activities will be described in following diagrams, others have not yet been understood and will need further research to properly interpret the data.