70

the COM in the lateral dimension. Figure 4.13 shows the resulting Q_i^*for such an attempt based

on the walk of Figure 4.11, which exhibits tightrope walking. A lateral Q^d
component of ±0.03 is

used instead of 0.0. The sign of Q^d
is such that foot placement target is away from the COM.

The resulting walk does not fall and exhibits much more realistic lateral foot placement.

4. 3

Stance-COM Regulation Variables

While using the up-vector and swing-COM RVs achieves

reasonable

success

in

generating

walking motions, similar use of the stance-COM RVs does not. Most trials fall during the second

controlled step. The failures are due to the fact that even modest changes in the stance-COM RVs

require very large stance hip perturbations which cause very large changes in the biped's state.

Figure 4.14 illustrates this idea. Because of this result, the chosen balance parameters fail to reach

a suitable state from which to begin the next cycle, even though the desired RV value might be

attained.

[!]Q^fwdsmall

[!]K large

While the walks all fall quickly, two interesting observations can be made:

1. The first controlled step reaches Q^dwith a high degree of accuracy, i.e. the discrete