*7.1. Test 1: Step Inputs*

This test consists of three separated step-waves with a duration of 8 seconds, each independently activating the neck tendons, as shown in Figure 11. The aim of this test was to validate the models' capability of responding to a static input. Since the dataset for the training and NN which corresponds to the FSS lacks individual tendon actuation, some error in the models is expected. Figures 12 and 13 show the outputs.

**Figure 11.** Test 1 input signal.

**Figure 12.** Test 1 results for pitch.

**Figure 13.** Test 1 results for roll.

Set membership scored 87.6260% in the NRMSE test for pitch and 79.3195% for roll. NN followed the output better and scored 97.9574% for pitch and 98.5654% for roll. The SM response did not settle at the appropriated inclination, lagging out before the maximum inclinations were reached. As can be seen, the model does not reach the negative inclinations properly. However, in Figure 12, the dynamics for negative and positive inclinations are followed as desired. It is seen that the existing bias in the time windows [3, 5] s follows the output dynamics. In the roll axis case, the same bias appears with the negative inclination angles, while in the positive case, it stabilizes at time 17 s in the final value. However, it does not properly capture that dynamic, neither. The maximum errors for the NN appear at 8 and 16 seconds, when the signal is changed from one tendon to the next; this is probably due to a lack of information in the data set. Meanwhile, RLS scored 79.9230% in pitch and 76.0284% in roll. This can be due to the fact that the initial conditions do not match the real model ones.
