Skip to content

Latest commit

 

History

History
105 lines (73 loc) · 4.15 KB

README.md

File metadata and controls

105 lines (73 loc) · 4.15 KB
Raw

ARGoS3-Construction

Simulation of collective construction for ARGoS3

This project consists of a simulation using ARGoS (see folder ''BuildingSim'') and a genetic algorithm using the MOEA-framework (see folder ''PopCodeGA''').

Configuration

Simulation

The simulation is configured via a '.argos'-file. The provided file can be found at ''BuildingSim/experiments/construction.argos'' This file enables the GUI. Another config file ''constuction-nsga2.argos'' disables the GUI and is used for the genetic algorithm.

You can configure the number of cylinders, arena size and target area with this file. For more details please see the official documentation

Genetic algorithm

Details about the genetic algorithm can be configured with a ''GA.properties'' file. A default file is located at ''BuildingSim/GA.properties''. You have to start the simulation from the same folder as the config file!

If no settings file is provided, default values are used for each parameter: The following parameters can be configured:

  • numRobots=10
  • PopulationSize=100
  • generations=1000
  • evaluations=3: Number of simulations executed during evaluation
  • tournamentSize=2
  • crossoverProb=1
  • mutationProb=0.01
  • gType=FULLCOUNT
  • resultPath=~/ArgosResults/ : Final log location

Requirenments

  • Maven
  • Java 8

Compilation

$ mvn package

Run

Genetic Algorithm

Using the script:

  1. chmod +x [gitroot]/scripts/run.sh
  2. ./[gitroot]/scripts/run.sh

or

Direct:

  1. $ cd [gitroot]/BuildingSim
  2. $ java -Djava.library.path=build/embedding/nsga2 -jar ../PopCodeGA/target/PopCodeGA-1.0-SNAPSHOT-jar-with-dependencies.jar

Single Run

  1. cd [gitroot]/BuildingSim
  2. ./build/loop_functions/construction_loop_functions/argosControl

For use with a specific genome, change it in the ''/BuildingSim/loop_functions/construction_loop_functions/argosControl.cpp''

or

Direct with unconfigured genome. See also ARGoS3-experiments for more details:

  1. $ cd [gitroot]/BuildingSim
  2. $ argos3 -c experiments/construction.argos

Logs

During a evolutionary run the following temporary files are created. All files have the same prefix. This prefix is the hostname if the run script is used, or "argos" otherwirse:

  • [prefix]_stats.txt: Information about this run. Usefull for many different runs
  • [prefix]_checkpoint.dat: A checkpoint file that is used if a run was interrupted. It is automatically deleted after a succesfull run. Delete this if you want to completely restart a run.
  • [prefix]_Solutions.dat: Binary file created by the MOEA framework. Use this to load the final solutions into MOEA again for analysis.
  • [prefix]_Objectives.dat: Objectives of all final solutions
  • [prefix]_log.txt: A extensive log about solutions in every generation

All files are generated in the same folder the programm was started in. After the algorithm is finished, the files are copied to the ''resultPath'' specified in the config file.

Console ouput

At start the current settings are printed.

The console prints solutionnumber, cylinderCoverage, sparsity, (accumulated Light) and the current genome of each solution that is evaluated. Additionally the coverage of each simulation during evaluation is printed. How many evaluations per solution are done, is configured with the "evaluations" parameter.

There will be ''generations'' * ''populationSize'' many solutions, each with its own line.

Similar information is written to the log.txt

log.txt

In the [prefix]_log.txt the following data is safed in csv format. Each line represents a solution with the following data:

  1. last evaluation number of the generation: divide by ''populationSize'' to get the generation
  2. hypervolume: the hypervolume of the generation
  3. obejctive: Cylinder coverage
  4. (objective: accumulated light): only used in the ''lightPeformance'' branch
  5. objective: sparsity of the swarm
  6. Genome: the genome used for the robots. You can copy-paste this into the ''argosControl.cpp'' to see the swarm in action.

Objectives are negated because the MOEA algorithms are minimizing.