Artificial neural network topology
This network topology has only three input variables to the left, the other
three are now weights of the last hidden layer. On the top view it can be seen that
there are three independent pathways that process the position, each of these correspond
to directions of the intended leg direction {dx, dy, dz}. The total number of
interneurons in this example is 12.
ANN performance comparison between tested topologies
Traditional network topology is shown by the solid line marked with (*); the distributed
network topology is marked with (o). Left: Deviation resulting after training was
stopped at 4 different stages. Right: When the average speed is moved away from
that used for training the performance of the traditional network deteriorates.