We present on this repository the code used in the paper AutoMoDe-Nata
to generate a repertoire of swarm robotics behaviors (here, for the e-puck robot),
and the AutoMoDe flavor Nata that uses this repertoire to automatically assemble
finite-state machines. Note that the repertoire can be automatically generated
or not, this makes this code compatible with AutoMoDe-Arlequin.
├── `Nata` this folder contains all sources corresponding to `AutoMoDe-Nata`
│ ├── `bin` this empty folder will contain the executable files.
│ ├── `env_rep.path` example file that contains the full path for the repertoire to use
│ ├── `gen` folder that contains the repertoire of `Arlequin`
│ ├── `missions` missions xml files for argos3 simulator
│ ├── `optimization` scenarios to use with irace to generate the robot control software
│ └── `src` source code of `Nata`
├── `Repertoire_generator`
│ ├── `generator`
│ │ ├── `bin` this empty folder will contain the executable files.
│ │ ├── `experiments` argos3 xml configuration files for repertoire creation
│ ├── `gen` this folder can store genomes
│ │ ├── `misc` this folder contains various script used to launch experiments
│ │ ├── `optimization` scenarios to use with irace to generate the robot control software
│ │ ├── `params` this folder contains the hyperparameter files for the generation of the repertoires
│ │ ├── `src` the source code of the repertoire generator
│ │ └── `startgen` the starting genomes for the repertoire generation
│ └── `results`
│ ├── `designRep.toml`
│ └── `repertoire`
└── `scripts`
- ARGoS3 (3.0.0-beta48)
- argos3-epuck (v48)
- experiments-loop-functions (repertoire)
- demiurge-epuck-dao (master)
First install all system dependencies:
sudo apt install cmake libfreeimage-dev libfreeimageplus-dev \
qt5-default freeglut3-dev libxi-dev libxmu-dev \
doxygen graphviz graphviz-dev asciidoc git \
libboost-all-dev libeigen3-dev openmpi-bin openmpi-common libtbb-dev python3-pip
pip3 install cppyy
We provide an automatic script (tested under Ubuntu 18.04) to install all software and their dependencies.
In this folder navigate to scripts and install all code repositories:
cd scripts
./bundle_argos3_epuck.sh
This should install the whole argos3 distribution and librairies in /home//argos-dist/. After the installation, if everything went smoothly, setup your $PATH for argos3:
./home/<youruser>/argos3-dist/bin/setup_argos
This command should be run after each login on the computer, you can add it to your .bashrc to make it automatic.
This command will also create the config.py file that contains informations about the path where ARGoS will find for the librairies and tools used for this experiment.
In order for the simulator to locate the repertoire in use, a file named
env.rep should be created in the current working directory where the
simulator is launched.
This file should contain the absolute path to the repertoire to be used.
For example for Nata:
<path_to_this_repo>/Repertoire_generator/results/repertoire/rep7_500g_d7-0/
After installation, (given config.py is correctly setup), if you wish to run an individual control software, it is done using the launchArgos.py script:
usage: launchArgos.py [-h] [--config CONFIG] [--gen GEN] [--fsm-config FSM_CONFIG] [--seed SEED]
Launch argos with options and automatic path resolving
optional arguments:
-h, --help show this help message and exit
--config CONFIG, -c CONFIG argos configuration file (.argos)
--gen GEN, -g GEN genome file (if launching Neuro evolution method)
--fsm-config FSM_CONFIG, -f FSM_CONFIG fsm descrption (if launching automode method)
--seed SEED, -s SEED
Example for Nata:
python3 scripts/launchArgos.py -s 98128 -c Nata/missions/foraging.argos --fsm-config " "
Example for Evostick:
python3 scripts/launchArgos.py -s 98128 -c Nata/missions/foraging.argos -g " "
Examples of genomes and PFSM that can be used with those scripts can be found
in the Controllers.zip file associated with the research paper and published
as IEEE Dataport
Using an existing repertoire to generate finite-state machines for a mission.
For running the optimization process, you need to have R with the irace package. To install R and irace:
sudo apt install r-base
wget https://github.com/MLopez-Ibanez/irace/archive/master.zip
R CMD INSTALL master.zip
After installation, set your $PATH for using irace:
export IRACE_HOME=~/R/irace/ # Path given by system.file(package="irace")
export PATH=${IRACE_HOME}/bin/:$PATH
Navigate to the optimization folder of Nata and launch irace:
cd AutoMoDe-Nata/Nata/optimization
irace --exec-dir=execdir --parallel 10 --max-experiments 50000 --scenario scenario.txt
This will launch a design phase in the current folder.
For the algorithm to be able to locate the repertoire of behaviors to be used,
you should create a env.rep file containing the path of the repertoire.
How to create a new repertoire of behaviors to be used to generate finite-state machines with the previous procedure.
/!\ This process requires a CPU cluster to be run in a acceptable amount of time. The scripts provided require modification according to the specific cluster computing engine that is used.
python misc/paralell_run_mpi_local_rep.py -d repertoire -j rep_500g -n 48 -e ../src/experiments/repertoireTrainingCluster.argos -p ../src/params/EvostickRepertoireParams500.ne -s ../src/startgen/evostickstartgenesRM11 -pr ../src/params/repertoiresParams7.rep -r 1
All the software prerequisite have to be installed before hand.
Clone this repository
Navigate to the repository and build the package AutoMoDe-Nata
mkdir build
cd build
cmake .. && make
Then, from the root of the repository, navigate to the repertoire generator and build it
mkdir build
cd build
cmake .. && make