Figure 4.

The effect of adaptive control on a single-joint lego robot (Figure 2). Each run uses a randomly generated desired trajectory q_d(t) over 20 s, and root-mean-squared-error is computed over the last 10 s only. The adaptive algorithm provides a significant improvement (p < 0.05; two-tailed t-test; 50 samples). Scatterplots show individual runs (with random jitter on the x-axis to avoid overlap), the shaded area is the mean plus or minus the standard deviation, and the 95% confidence interval of the mean is shown.