Generated trajectory simplex trained recurrent network experiment figure

Figure 12a. Desired trajectory of Experiment

simulations with the initial state of all the neurons set randomly according to N(100,5) were conducted. Again, all the results obtained demonstrate convergence of the network to a single circular trajectory. One example of the various simulations is shown in Figs. 12a-c and 13a-b. Figs. 13a and b show the outputs of the two output neurons which demonstrate that the recurrent network has indeed captured the oscillating behavior required to generate the desired circular trajectory. Figs. 12a-c show the desired trajectory, the trajectory produced by the network, and both trajectories overlaid, which confirm that the

© 2001 by CRC Press LLC

Figure 13b. Desired and generated y-coordinate output of a trained recurrent network (Experiment 1)

2. EXPERIMENT 2