Figure 6. Progress of the neural learning
Observing the gait behavior of the robots trained us- ing neural networks, they are much more stable and ro- bust than the gait behavior when controlled directly by the half-ellipse, because the neural network uses sensor information (bumpers) and have the capacity to general- ize to non expected situations. The Figure 7 shows the gait control generated using locus based gait, and The Figure 8 shows the gait control generated using the neu- ral network3 .
Figure 7. Half-ellipse robot gait