Genesaver.md #1

{{ Project | Genesaver |

`  curator  = Sam Stafford                    |`\
`  path     = //guest/sam_stafford/genesaver/ |`\
`  download = //guest/sam_stafford/genesaver/bin.ntx86/ |`\
`  language = C/C++ |`\
`  license  = GPL   }}`

### About This Project

Genesaver is a small virtual ecosystem in the form of a screensaver. The
"creatures" are steered around the screen by simple neural nets.
Successful creatures produce offspring that inherit modified versions of
their parent's behavior, so that increasingly intelligent behavior
evolves over time.

#### Installation and Configuration (Windows)

Save , right-click, and Install. On older versions of Windows, save
directly in your C:\\Windows\\System32 folder.

The "Configure" option will open Notepad with the **Genesaver.cfg**
configuration file. Edit options as desired, save, and exit. The options
you are most likely to want to modify are:

-   **plantgrow** - higher values will support a higher creature
    population, which will also increase CPU load
-   **crtrscale** - smaller values make creatures smaller, effectively
    making the world larger
-   **zoomprcnt** - you may want to increase this if you decrease the
    creature size
-   **blurtrail** - this is pretty, but may not work well on all OpenGL
    implementations

#### The Genesaver Ecosystem

Red preys on green. Green preys on blue. Blue preys on red. The grey
dots are plants, which spawn at a constant rate.

![](Genesaver.jpg "Genesaver.jpg")

When a creature collides with something it can eat, it takes a "bite"
out of it, reducing the prey's energy by some amount and increasing its
own energy by a fraction of that amount. Individual creatures may be
more adapted to eating either plants or other creatures; this is visible
by the color of the creature's tail, with black tails indicating
herbivores (able to take big bites out of plants) and white tails
indicating carnivores (able to take big bites out of creatures). This is
on a sliding scale, so some creatures may be omnivores (gray tails) and
be able to take medium bites out of either.

Creatures continually burn energy, and they burn it more quickly as they
move. When a creature runs out of energy, it dies and vanishes from the
world. If a creature reaches 200% of its starting energy, it reproduces
asexually, putting half of its energy into a new creature with a
modified version of its own genome. If two creatures of the same color
are at 150% of their starting energy and they collide, they reproduce
sexually, producing an offspring with a mixture of genes from both
parents; when this happens their tails flash yellow.

When carnivores of a particular color are overpopulated relative to
their prey population, all creatures of that color will begin damaging
each other as they compete for food and try to restore equilibrium; when
this happens their tails flash purple.

#### Creature Behavior

Each creature's behavior is guided by a simple neural net. Most of the
input neurons provide information about nearby creatures/objects of
different colors; others provide information about the creature's own
energy level and whether it's currently overpopulated.

The input values are fed to two levels of intermediate neurons through a
series of weighted connections before reaching the output neurons, which
control the creature's movement.

The "chase" camera mode (press C to cycle camera modes when the
screensaver is running) shows a view of a single creature that includes
its neural net:

![](Genesaver braincam.jpg "Genesaver braincam.jpg")

In the above screenshot of a green carnivore chasing blue prey, input
neuron A has a positive (blue) value because the creature's left eye
sees something blue. This signal feeds through intermediate neurons C
and D and results in a positive value for output neuron F, which makes
the creature turn left.

Input neuron B's value reflects the creature's current energy, and is
connected to output neuron E, which controls forward motion; hence this
creature will move forward in search of more prey while its energy is
high, and might slow or stop in order to conserve energy when it's low.

Creature behavior is not in any way "hard-coded" into Genesaver; the
first time you run the screensaver, every creature's synapses are
randomized, but they will evolve over time.

#### Creature Genetics

Each creature has its own unique genotype, which determine its:

-   Color
-   Diet
-   Synapses

Synaptic connections determine a creature's responses to its
environment. An individual creature's neural net will not change over
its lifetime; hence individual creatures do not learn.

When a creature reproduces, the offspring receives a copy of its
genotype, with a small percentage of the genes modified randomly. Since
creature behavior is genetic, a "family" of creatures will "learn" over
multiple generations as certain genes connect neurons to produce
behaviors that lead to reproductive success.

When Genesaver exits, it will save all of the current creatures to a
file called **Genes.txt** so that they can be restored the next time the
screensaver starts.

#### History

I wrote Genesaver as a hobby project while I was a student at Cal; it's
the oldest app I've written that I continue to use today. I had just
taken an introductory graphics course, which inspired me to use OpenGL
to make a simple screensaver, but I had not yet done any user interface
work, everything I knew about neural nets and genetic algorithms had
come from my own casual reading, and C++ still seemed like a strange
foreign language. The inelegance of both the code and the interface is a
testament to the author's inexperience.

It's my favorite screensaver ever.

{{ License | GPL | 2008 | Sam Stafford }}

{{ RecentChanges | //guest/sam\_stafford/genesaver/src/ }}

<Category:Screensavers>