Whilethe

6. 1. 5

Automatic Synthesis

Another possible direction for future work is toward automatic controller synthesis.

base PCGs, RVs and LPPs for the motions given throughout this thesis were designed by hand,

they should ideally be automatically generated.

Previous use of pose control has been directed

primarily toward such automatic synthesis of motion controllers [vKF94] [vKF94b] [vL95].

Most of these attempts deal with relatively simple, statically-stable creatures.

automatic synthesis of control for statically unstable 3D systems is a difficult one.

The problem of

In [vL95], van

de Panne and Lamouret propose the use of external guiding forces to initially generate cyclic

PCGs, followed by two subsequent phases which reduce the forces and then attempt to eliminate

them respectively. While the motions generated by the second phase are significantly more natural

looking than the initial PCGs, they are not fully realistic in the sense that the creature's actuators

do not drive the motion unassisted.

The final phase, removal of external forces, is not always

successful and can be very computationally expensive.

Rather than attempting to directly synthesize the entire PCG in this way, it might be possible to

automatically synthesize various control components for our system.

candidates are the base PCG and the fixed PCG perturbations (LPPs).

The

two

most

likely

Base PCGs for various

interesting motions might be synthesized using van de Panne's first two phases and then stabilized

using the balance control presented in this thesis.

LPPs might be synthesized using a more

traditional generate, test, and refine process such as that used for simple, statically stable systems.