GA report
Date: May 22, 2022
Abstract
This simulation project is aiming on evolving random noise into creatures in a 2-d virtual environment
while trying to minimum human’s predene and intervention. Genetic algorithm is used to generate
and evolve the brain’s neural system of the creature, which controls all the individual’s behaviors.
Creatures are allowed to interact with environment and each other, which leads to the natural selection
to happen. After evolving for thousands of generations, creatures shows a certain level of intelligence
to solve simple problems. Similar behavior patterns also been found between virtual creatures group
and the real animals, like escaping from predators and seeking for food. The evolutionary process of
the virtual world also matches the fossil record for evolution in early stage of earth.
Keywords:
1 Introduction
2 Related Works
3 Methods
3.1 Virtual Environment
64 × 64
1 × 1
3.2 Creature Control
3.2.1 Brain Structure
Sensor
Inner Neural
Activator
Link Rule
tanh(
∑
inputsignals)
3.2.2 Represent Brain
Figure 1:
3.3 Selection Strategy
3.3.1 Synchronous
3.3.2 Asynchronous
3.3.3 Adversarial Selection
3.4 Reproduction and Mutation
4 Results
4.1 Solve Simple Problems
4.2 Cell Behavior in Adversarial Simulation
Initial Stage
Ediacaran Period
Middle Stage
Stable Stage
Figure 2:
5 conclusion
6 Future Work
References
Primerpython https://github.com/Helpsypoo/primerpython
biosim4 https : / / github . com / davidrmiller / biosim4
The Life Engine https://github.com/MaxRobinsonTheGreat/LifeEngine
Proceedings of the 21st annual conference on Computer graphics and inter-
active techniques - SIGGRAPH ’94 10.1145/192161.192167
