(Emergent Morphology)

8 Results

While hopes for Olympic performance were somewhat dampened, there was no lack of amusement. Improved locomotion was clearly achieved by all populations, but often at the expense of the head, which in some populations was swung wildly to shift the center of mass, or used like a foot. In all cases, the fitness reward for locomotion was set at a constant weight, while the secondary head-related fitness terms were set at differing weights, for purposes of experimentation. Choosing the best proportions of weights for multiple fitness terms is not trivial, as indicated also in experiments by Fukunaga et al., (94). In these experiments, it is highly dependent on the degrees of freedom in the animat, the duration of the evaluation, and the nature of the physical model.

It has been found that by merely discouraging rapid head movement and collisions via fitness penalties, other body parts of evolving animats will tend to function not only for locomotion, but also to keep the head steady. In tests where animats were rewarded for holding the head up higher, the result was more upright postures, and some pseudo-human forms - but with one distinctive non-human feature - a magnificent kangaroo-like tail which helped balance the animat, tripod-style, as it awkwardly ambled along. Without any stimulus/response system modeled, these animats have no sense of balance so as to adjust their motions accordingly, thus, sustained bipedal walking is near impossible. Figure 3 shows two such animats from populations which evolved through fitness pressures for head height.


Figure 3 Two animats from populations which have evolved through fitness pressures for head height.

The fitness pressures for discouraging head movement and head collisions had to be fine tuned a number of times before satisfactory results came about. If a fitness term's weight was set too high, such as the penalty for head movement, locomotion was inhibited - since locomotion involves at least some head motion - and the population converged on immobility. Once reasonable ranges for these settings were established, desired behavior was more easily achieved.

Resulting gaits were of a large variety. Figure 4 shows the motions of an animat representing a population which evolved with no fitness constraints on the head, and so each animat in the population was free to do with its head whatever was necessary for locomotion. This animat demonstrates the use of the head as a third foot, aiding in balance.

Figure 5 shows an animat representing a population which evolved through pressures for holding the head high. This animat is roughly horse-shaped, except that it has six legs, and it moves like a crab - sideways.

Alternating limb motions emerged in many populations (as in Figures 4 and 5) but quite frequently populations converged on gaits in which all the limbs swung in unison, creating hopping, inching, and other such strategies. Figure 6 shows an animat which moves in this manner. Of note is the fact that this animat moves backwards (the direction of movement is not the direction in which the head is aimed). This illustrates the fact that the head is not used to lead motion, but only to modify it.


Figure 4 An animat representing a population which evolved with no fitness pressures pertaining to the head. The head is used as a foot. In this depiction (and all which follow), a series of after-images are shown in each frame to indicate the rates and directions of motion in each body part. The arrow indicates direction of travel.



Figure 5 An animat representing a population which evolved with fitness pressures for holding the head high.
It might be described as a six-legged horse which moves like a crab - sideways.



Figure 6 An animat representing a population which has evolved with slight constraints on head motion.
This animat does not travel in the direction that the head is aimed, indicating that the head is not used to lead motion, only to modify it.

9 Interactive Evolution Overlay

(go to beginning of document)