85

torso rotation causes the biped to rotate about its own vertical axis more quickly than it can move

along a circular path. As a result, it turns to face the centre of the circle.

In the F-L and L-F sampling trials, the highest turning rates eventually cause the biped to turn

around completely and walk almost straight backwards. The turning perturbation is ineffective in

the backward walking cases because the model walks primarily on its heels.

never obtains sufficient ground reaction forces to fix the foot's orientation.

As a result, the foot

(a) - F-L sampling

(b) - L-F sampling

(c) - SP sampling

Figure 5.7- Hip plots for the most successful
over a range of scaling factors. Q^{dturning perturbation trials}
= [.25,0] for all trials.

Not all values of Q^dwork equally well as Figure 5.8demonstrates.

Linearly

varying

the

perturbation scaling factor still yields smooth behaviour, but not the desired results of Figure 5.7.

While this result makes it clear that some care must be taken in choosing parameters, it also serves

to illustrate the fact that reasonable choices can often be made without exploring a large parameter