Evolution of Flight

Method

I created a program that would generate 100 random airplanes based on thirteen different ‘genes’, shown below.

Here is the first generation of airplanes. As you can see, the thirteen genes are able to create a wide variety of forms.

I selected ten that I thought would fly the furthest (marked in red), lasercut them, then tested each one. My test setup consisted of  a stand that marked the precise angle (50 degrees) and height (60 inches) that I would drop each plane from. I flew each plane three times to ensure for consistency. My goal was to have a plane fly 120″, achieving a flight ratio of 2:1.

Next, I had to create a new set of planes based on the previous generation’s results. My algorithm selected two planes and crossbred them to create a new plane. The further the plane flew, the greater chance it had of being selected.  This mirrors the same method that sexual organisms use for reproduction, where the offspring contains some genes from parent A and some from parent B, resulting in a brand new design. A hundred new designs were created for each generation. I then repeated the tests with each generation. Below you can how the design of the planes progressed over time.

Results

The graph of the average flight distances shows an interesting progression. The improvement in flight distance is non-linear; it jumps up at certain generations.

These changes reflect the process I used to select the planes. Initially, I was selecting planes based on what was interesting or what I thought would fly well. After generation 4, I began to observe the properties of the planes that flew further and used my findings to select planes for the future generations. This led to a large improvement in the average flight of each plane and I was quickly able to narrow down the features of a successful airplane. Surprisingly, a lot of the properties that were successful weren’t ones I would have picked initially. Using an evolutionary algorithm helped unearth interesting new solution, such as a spinning airplane, that weren’t obvious to a designer.

I was also able to see that although some designs flew consistently well, they didn’t have as much potential as other designs that flew farther but less consistently. Without human interference, the less consistent designs may have taken hundreds of generations to develop and surpass the more consistent planes.

Evolution of Flight begins to suggest how the process of evolution can be improved with human intuition and also how the human design process can be augmented using evolution as the creative driving force. I believe that we will start seeing the design process use evolution more and more as we begin to understand the right balance between computational creativity and human intuition.

Full Process

– Intro

– Slots and Planes

– First Batch

– Testing

– Testing 2