MLE project

The MLE model

Beurier G., Simonin O., Ferber J. (2002, December). Model and simulation of multi-level emergence. In 2nd ieee international symposium on signal processing and information technology, isspit’02, pp. 231–236. Marrakesh, Morocco.

In the preceding reference, a model of multi-level emergence (MLE) of complex structures is defined using a unique and very simple recursive agent behavior. More precisely, starting with only one kind of agents (level-0), the agents evolve and build a recursive structure having a circular shape. That is, level-0 agents turn around level-1 agents that turn around level-2 agents and so on.

To achieve this, the agent behavior relies on perceiving, emitting and reacting to three different types of pheromones: (1) presence, (2) repulsion and (3) attraction. Presence is used by an agent to evaluate how many agents are in its vicinity and decide if it has to mutate to the next or preceding level. Roughly a mutation occurs if an area is overcrowded or empty. Repulsion and attraction are both used by the agents to create a circular zone of attraction around them. The behavioral process is decomposed in four stages: Perception, Emission, Mutation and Move. So, the state and the behavior of each agent is completely defined by only one integer which is its current level. The level simply modifies the agent’s emission rate and limits perceptions to pheromone of adjacent levels.

Objectives

The MLE project is about simulating and using the previous model for benchmarking the integration of GPGPU in the third version of TurtleKit, a multi-agent based simulation platform :

MLE Demonstration

Use the following button to launch the MLE simulation without any installation using Java web start.

If you cannot use Java Web Start for any reason, you can download the following executable jar file : MLE.jar and execute it by double clicking it, or running it using a command line : java -jar MLE.jar

For benefiting from the Cuda GPU boost, under Linux you should install the Cuda driver and the Cuda toolkit 5.5 from Nvidia. Windows users can download the source code on the TurtleKit web site and manually compile the Cuda kernel which is in the cuda package.

Demo videos

First video: MLE Model without GPU, Environment : 512 x 512 -> 262144 cells, Population : 10 % -> 26215 agents.

Second video: MLE Model with GPU, Environment : 512 x 512 -> 262144 cells, Population : 10 % -> 26215 agents and then 20 % -> 52429 agents.

TurtleKit software

Test online

Download

Latest TurtleKit version

This software requires a Java runtime environment 1.7

Content

This file contains binaries, the TurtleKit public API, and the source code of the platform. TurtleKit is designed to be a standalone library packaged as a jar file. So, you can start using it by adding it to the class path of your own projects just like you do with any Java library. Project samples are available in the demo directory of the archive.

Installation

Installation : unzip in some directory and use the jar file as a regular Java library in your own projects. Also, the jar file is runnable and will launch the TurtleKit desktop with some demos.