Framsticks

This page summarizes basic experiment definitions that are available within Framsticks. Each section presents one experiment definition along with its properties and short descriptions. All these settings are available in the Framsticks GUI and for scripts.

Jump to experiment definition:

• Evolutionary optimization (steady-state)
• Evolution (endogenous fitness)
• Demonstration of evolution
• Neuroanalysis
• Learn where food is, explore and exploit
• Boids
• Evolutionary optimization (generational)
• Walking in a maze
• Text writer
• Dance
• Framsbots - Game
• Multiple evaluation of loaded genotypes

Experiment definition: Evolutionary optimization (steady-state)    (short name: standard)

This experiment definition can be used to perform a range of common experiments. It provides

– one gene pool for genotypes
– one population for simulated individuals
– one "population" for food
– steady–state evolutionary optimization
– fitness as a weighted sum of performance values
– custom fitness formulas
– fitness scaling
– selection: roulette or tournament
– multiple evaluation option, average and standard deviation available
– can produce logs with average and best fitness
– can detect stagnation and stop automatically, or start the second fast–convergence optimization phase
– can save best genotypes

This experiment definition has 44 properties – see table below:

Experiment definition: Evolution (endogenous fitness)    (short name: reproduction)

Evolution in a population with asexual reproduction.
Each creature with its energy level above threshold produces a mutant offspring. The offspring is born near its parent and receives half of parent's energy.
Food is created at a constant rate and placed randomly.

This experiment definition has 13 properties – see table below:

Experiment definition: Demonstration of evolution    (short name: evolution_demo)

This experiment definition demonstrates the process of evolution.
Individuals are placed in a circle. A new individual is cloned, mutated, or crossed over.
Then it is evaluated in the middle of the circle, and – depending on its fitness – may replace a poorer, existing individual, or disappear.

This experiment definition has 35 properties – see table below:

Experiment definition: Neuroanalysis    (short name: neuroanalysis)

This experiment evaluates all genotypes in the gene pool. During simulation, the output signal of each neuron is analyzed, and its average and standard deviation are computed. These data are then saved in the 'Info' field of the genotype.

This experiment definition has 3 properties – see table below:

Experiment definition: Learn where food is, explore and exploit    (short name: learn_food)

When an individual encounters food, it eats a bit of it and remembers its location. It also gets "refreshed" (i.e. gets a full amount of energy). Energy of each individual provides information on how recent (current) is the food location information that the individual stores. Information recency is visualized as brightness of individuals (light green ones have recently found food). When individuals collide, they learn from each other where food is (by averaging their knowledge). A newborn individual moves randomly and receives (duplicates) knowledge from the first knowledge–rich individual that it collides with. An individual that cannot find food for a long period of time dies, and a newborn one is created.

–=–

An interesting phenomenon to be observed in this experiment is how sharing information helps explore food location. Food items can be added either close to previous items, or randomly (in the latter case, the information about food location is not very useful for individuals). You can turn off automatic feeding and keep adding food manually to see how learning influences behavior of the population. See the "share knowledge" parameter (on/off).

With learning, individuals do not have to find food themselves. They can also get in contact with other individuals that know where the food was, and exchange information (the values learned are proportional to the recency of information). It is interesting to see how knowledge sharing (cooperation, dependence) versus no sharing (being self–sufficient, independence, risk) influences minimal, average and maximal life span in the neighboring and random food placement scenarios.

Notions of exploration of the environment and exploitation of knowledge about the environment are illustrated as well in this experiment.

–=–

The dynamics of this system depends on the following parameters:
– number of individuals and world size
– size and shape of their body (affects collisions and sharing of knowledge)
– food eating rate
– food placement (neighboring or random)
– learning strategy (e.g. weighted averaging of food coordinates)
– behavior (e.g. move within circles, small after finding food, then larger and larger)

–=–

Technical:
Food location (x,y) is stored in data–>food of each individual (as an XYZ object, z coordinate is ignored)

This experiment definition has 11 properties – see table below:

Experiment definition: Boids    (short name: boids)

Boids, developed by Craig Reynolds in 1986, is an artificial life program, simulating the flocking behaviour of birds.

As with most artificial life simulations, Boids is an example of emergent behaviour; that is, the complexity of Boids arises from the interaction of individual agents (the boids, in this case) adhering to a set of simple rules. The rules applied in the simplest Boids world are as follows:

* separation: steer to avoid crowding local flockmates,
* alignment: steer towards the average heading of local flockmates,
* cohesion: steer to move toward the average position of local flockmates.

(http://en.wikipedia.org/wiki/Boids)

This experiment definition has 9 properties – see table below:

Experiment definition: Evolutionary optimization (generational)    (short name: generational)

A simple "genetic algorithm" experiment:

– two gene pools (previous and current generation)
– one population for individuals
– generational replacement of genotypes
– selection: roulette, random, tournament
– fitness formula entered directly into the group's field

This experiment definition has 9 properties – see table below:

Experiment definition: Walking in a maze    (short name: mazes)

This experiment definition can be used to evaluate (and evolve) creatures moving between two specified points in a maze. These points are indicated by start and target marks (in the second population).

Genotype's user1 field (which acts as a maximized fitness) contains
– energy left (when target found during lifespan)
– distance to the target (as a negative value) when a creature died away from the target.

Press "SHIFT" and click the right mouse button to manually set the start and finish points.

Use maze[1,2].sim settings for this experiment definition.

This experiment definition has 39 properties – see table below:

Experiment definition: Text writer    (short name: text_writer)

This experiment definition displays formatted (and flowing) text using creatures as letters and digits. It requires the "fonts.gen" file. Only big letters and digits can be used in the text.

Initialize the experiment to situate the text.

The text can be formatted using HTML–like tags:

<left>
<center>
<right>
<justify> – are used to start a paragraph

<hNUMBER> – sets the height (elevation) of the text (see also the 'Gravity' setting)

<hsNUMBER> – sets hotizontal spacing between letters

<vsNUMBER> – sets vertical spacing between lines

<f0>
<f1> – select the font (f0 is more regular)

This experiment definition has 3 properties – see table below:

Experiment definition: Dance    (short name: dance)

A synchronous framsdance :–)

This experiment definition has 38 properties – see table below:

Experiment definition: Framsbots - Game    (short name: framsbots)

Framsbots is a simple game inspired by the classic Robots game.
The aim of this game is to run away from hostile creatures and make all of them hit each other.
Just click somewhere (left–click or double–right–click) to move your creature (the one that is in the middle of the world in the beggining). Your creature will go towards the point you clicked.
All the enemies will move towards you. Use this information to make them hit each other, so they will loose energy and die.
If you see an apple, try to get it. You will gain energy and you may even get a new life!
Use shift+left mouse drag to rotate world.

Read more about this game:
http://www.framsticks.com/wiki/FramsBots

This experiment definition has 16 properties – see table below:

Experiment definition: Multiple evaluation of loaded genotypes    (short name: standard-eval)

Use this experiment to evaluate all genotypes one by one.
Use the "Repeat count" parameter to set the required number of evaluations of each genotype.

First load your genotypes for evaluation, then define the evaluation plan, initialize the experiment, and finally run the simulation to perform all evaluations.

After evaluation, the selected key in the Genotype.data Dictionary ("evaluation" by default) contains a Dictionary of results. These results can also be saved to a file, as defined by the "evalsavefile" parameter. Note that for whatever reason (e.g. incorrect genotype), the dictionary you will get may be empty or partially empty and may not have the fields you expected, so handle such cases properly.

This experiment definition has 37 properties – see table below: