UNAL 2018_I - Artificial Life Project
Artificial environment developed as final project for Artificial life course at Universidad Nacional de Colombia
Modeled agents correspond to fishes. There are two species coexisting at the environment: preys and predators.
Both of them have features as
- Speed
- Metabolism rate
- Energy level (current): Increased by eating, decreased at metabolism rate
- Minimum energy level: If the energy level becomes less than this, the agent dies
- Vision range* (preys only)
- Skin* (preys only)
- Shape
Preys tend to need more food, since eating plants increase their energy level by a small value. Predators can become inactive for periods when they reach a food satisfaction threshold, such level is reached when eating approximately 2 fishes.
Note: Preys also have a food satisfaction threshold, when it is reached, boids and escape behaviors take control; anyway it is a ephemeral state and they'll need to get food soon.
Agents
Preys:
Predators:
Behavior
The agents' behavior is determined by simple rules.
Preys:
Several behaviors are mixed, according to certain precedence:
- Escape from predators
- Seek food
- Flock motion
- Cohesion
- Alignment
- Separation
Predators:
This behavior is a little bit simpler than preys':
- Hunt when the current energy level is under food satisfaction level
- Be quiet (reduced velocity) otherwise
Each agent loads a default image and performs and affine non linear-transformation which tries to mimic fisheye distortion
Examples:
Preys:
Predators:
Preys have skins which are made using Turing patterns - Reaction diffusion systems; based on Reaction-Diffusion Tutorial, by Karl Sims, where he explains in detail the differential equations and convolutions required.
Each skin correspond to a grid, where each cell keeps 2 substances (activator and inhibitor) at different levels, and its colors is defined depending on those quantities. Parameters such as feed or kill are set based on Gray-Scott Model. Available at: Reaction-Diffusion by the Gray-Scott Model: Pearson's Parametrization.
Java threads are used to generate a 100 x 100 image at execution time (for each prey agent).
Feed and kill constants are randomly chosen from some preset parameters.
The colors are randomly modified using 2 rgb colors as base:
- Green (200 * (activator level at the specific cell), 255, 0)
- Dark yellow (255 , 200*(inhibitor level at the specific cell), 28)
They are mixed giving slight preference to one or another
Examples:
These results are achieved through the variation -combination of the mentioned parameters.
Plants represent food sources for preys. They're build using L-Systems. Based on The Algorithmic Beauty of Plants Book
Food sources have several states which represent seasons (different growing stages depending on location). This feature lets us see interesting emergent behavior as migrations.
Evolution is achieved through the use of a genetic algorithm. Regularly a iteration is performed and the entire population is replaced using selection, reproduction, crossover and mutation principles.
- Selection: Some individuals are chosen depending on features such as: speed, metabolism rate and vision range. A tournament is performed and pairs of individuals are picked to reproduce.
- Crossover: Each pair of parent individuals give born a pair of children, which features (speed, metabolism, vision) are set based on their parents' values
- Mutation: The features of each new individual are modified, those values can be increased or decreased based on certain probability.
Java and Processing are required to run this program.
The structure of the project is made for running it on an IDE such as IntelliJ IDEA or eclipse.
The instructions to use Processing library on Eclipse can be found here
A video which shows the simulation working can be found here