Panspermia

Panspermia, from 1990, visualizes the theory that life in the universe exists in the form of seeds and spreads spores. The two-minute computer animation be­gins with a grain flying through the universe, which hits a planet and bursts. From the spores scattered into the surrounding field different plants evolve. In the animation sequences that follow individual species grow: ferns, winding plants and trees. In one sequence, the camera drives through a forest with di­verse plants (see Fig. 12.3). The animation ends with individual plants forming seeds, which they shoot out into space: the cycle of life.

Sims used procedural models for the production of the plants: 21 parameters affect the condition and growth of a three-dimensional hierarchical structure of connected segments. From the parameters of the procedural model, growth direction and growth rate as well as positions for new buds are computed. Sims does not present details of the generating algorithms; however, he refers to the work of Aono and Kunii [5], de Reffye et al. [34], as well as Viennot et al. [225] (see also Chap. 4). The resulting hierarchies of segments are fitted with cylinders, generalized cylinders, cones and polygons for the visual display. Furthermore he developed software that facilitates control of the evolution­ary process interactively. A selected genome is mutated, and the different mu­tations are either represented in a table as a wireframe model, or as high – resolution pictures, which can be switched through with the mouse. A selected

Figure 12.3

Stills from the Animation “Panspermia” (Karl Sims, 1990)

 

Panspermia

model can be stored and adapted manually for the animation in Panspermia. For example, colors and textures were later changed and added, and in some cases during a separate process evolved leaves were joined. The typical devel­opment of a plant requires between 5 and 20 generations.

Sims does not achieve the animated effect of the vegetation using highly de­tailed drawings of the individual plants, which could not have been obtained because of the mass of the plants shown, but through an organic movement using a dynamic simulation.

Подпись: genetic programmingIn [199] he systematizes his techniques by interconnecting parameter sets, which are applied in Panspermia. If the genome consist of a set of N param­eters, as is the case, for example, in procedural models, then one can regard the genome as a point in N-dimensional space. Two source genomes define the corner points G1 and G2 of an N-dimensional cube. A new genome can be recombined using two given genomes according to the following rules:

1. In a crossover, parameters from one of the two given genomes are grad­ually copied into the resulting genome. With a certain frequency the source genome alternates between the two given genomes. This has the effect that the neighboring parameters have a higher probability of orig­inating from the same source genome and, thus, characteristics that de­pend on several neighboring parameters will remain intact.

2. If for each parameter the source genome is selected randomly, and its pa­rameter is copied into the resulting genome, then the generated genome

is located in a comer point of the N-dimensional cube, similar to the first case. Local dependence is lost, and the variance is broader.

3. The two source genome vectors are linear interpolated: Gres = pGi + (1 – p) G2, p is selected randomly from [0… 1]. The new genome is on the connecting line between Gi and G2. A soft transition illustrated as an animation could be termed a “genetic dissolve”.

4. Each parameter is interpolated with a per parameter randomly selected p from the source genome parameters. The generated genome lies some­where within the cube spanned by Gi and G2. Thus, the variance is large, but the parameters, which work in both source genomes in opposition, such as left curvature versus right curvature, could interfere (although curvature is desired, the result might be straight).

PanspermiaFor Panspermia, Sims chose the second variant, since he aimed at variance, and at keeping the model characteristics unchanged without the influence of local neighborhoods.