This project is a fork of https://sourceforge.net/projects/polyworld/. The purpose of this fork is to allow for experimentation with emergence of structured communication. Any useful results will hopefully be pushed back to the original project.

Ubuntu 12.04 or higher

Instructions in "code boxes" are to be executed in a terminal.

If you don't already have git, install it via:

sudo apt-get install git

Download the source with the git command. This will create a directory named polyworld-comm under your current directory. If, for example, you execute this command from your home directory (/home/suzy), then the source will be in /home/suzy/polyworld-comm.

git clone https://github.com/sean-dougherty/polyworld-comm.git

Go to the polyworld-comm directory you just downloaded:

cd polyworld-comm

Polyworld requires that a number of packages be installed. If you're using a supported version of Ubuntu then you can simply do the following to install them:

scripts/install/packages-linux.sh

It's finally time to compile the source! Issue the following command:

make

If it worked then you should see a message like "scons: done building targets." and ./Polyworld should exist, which you can verify via:

ls ./Polyworld

Now, let's see if Polyworld runs on your machine. Execute the following:

./Polyworld worldfiles/hello_world.wf

You should see a graphical window appear containing a 3D simulation.

To download updates posted to Github by another author, you should first open a terminal and then change directories to where you originally downloaded the source. For example, if you originally downloaded to ~/polyworld-comm, then you would execute:

cd ~/polyworld-comm

Next, download the github content:

git pull origin master

Finally, you should do a build to make sure your binaries are current:

make

If you would like to use the genomes of agents evolved in some Simulation A as the first generation in a Simulation B, then you must:

If another author has already designed the worldfiles of A and B to perform an import, you can skip this step.

Simulation A must record its genomes and (optionally) the agent positions. To record genomes, you must have the following in A's worldfile:

RecordGenomes True

Or, you can make the simulation record everything, including genomes:

RecordAll True

If you would like to preserve the positions of the agents, which can be important in simulations that have distinct sub-populations, then you must have the following in A's worldfile:

RecordPosition Precise

You may also use the RecordAll property to cause the recording of positions. Note that there is also an Approximate setting for the position log, but that format is not currently supported by the seed code.

Like step 1, you may skip this if another author designed the worldfiles

Warning! The agents in Simulation B must have exactly the same genome schema as in Simulation A. For example, you can't add new internal groups or a new sensor to Simulation B.

The worldfile from Simulation B must be modified to specify it is importing its seed genomes from a run. To do so, you must have the following in Simulation B's worldfile:

SeedGenomeFromRun True

Next, you must specify the number of seed agents that will be created. For example, if you want 180 agents from Simulation A:

SeedAgents 180

If you want to preserve agent position information, then you must include the following in B's worldfile:

SeedPositionFromRun True

The run directory for Simulation A musn't be the default directory ./run. You must rename/move it.

If you are using worldfiles from github, it is suggested that you create a runs directory with a structure reflecting that of worldfiles. For example, if your Simulation A is from worldfile worldfiles/comm/predator/deaf/0.0-motionless-eating.wf, then use the following commands to move its run directory:

mkdir -p runs/comm/predator/deaf/0.0-motionless-eating

mv run runs/comm/predator/deaf/0.0-motionless-eating/run_0

Simulation B requires the presence of the file ./genomeSeeds.txt, which can be conveniently generated with the tool scripts/genomeSeed. The tool is capable of several behaviors, which you can learn about by running the command with no arguments. By default, it will generate a file for the final generation of agents in the run it is pointed at.

To generate a seedGenomes.txt for the final generation of Simulation A, assuming the run directory is located at runs/comm/predator/deaf/0.0-motionless-eating/run_0, execute the following:

./scripts/genomeSeed runs/comm/predator/deaf/0.0-motionless-eating/run_0

If would you like to also import agent positions, then you must specify the --pos flag:

./scripts/genomeSeed --pos runs/comm/predator/deaf/0.0-motionless-eating/run_0

You're now ready to execute Simulation B. Phew!

If you've done things correctly, you should see a large number of messages written to stdout when you execute Simulation B indicating that it is seeding from a previous run. An example line of output follows:

seeding agent #1 genome from runs/comm/predator/deaf/0.0-motionless-eating/run_0/genome/agents/genome_184850.txt

These instructions assume you are currently in the polyworld-comm directory in a terminal session and that the run you want to analyze is located at ./runs/comm/predator/hearing/1.0-collaborative-eating/run_0

Execute the following, which might take a long time:

./scripts/CalcSound ve ./runs/comm/predator/hearing/1.0-collaborative-eating/run_0

Execute the following:

./scripts/comm/plot_voice_entropy ./runs/comm/predator/hearing/1.0-collaborative-eating/run_0