This repo contains a HyperNEAT experiment to produce an activity potential field, known as the Shunting Model, in a reactive and deliberate task, known as the River Crossing Task (RC task).
The details of the RC task and the Shunting Model can be found in 'Neuroevolution of Agents Capable of Reactive and Deliberative Behaviours in Novel and Dynamic Environments' (see links section).
The paper associated with this work can be found below:
[1] Toward evolving robust, deliberate motion planning with HyperNEAT
This repo is an add-on to Jason Gauci's HyperNEAT v4.0 for C++; a running version of that code is needed.
An install.sh is provided and makes the appropriate changes to HyperNEAT v4.0 for the RC Task environment. Each experiment has a static environment number reference (see Hypercube_NEAT/include/HCUBE_Defines.h), this environment occupies number 30.
Note: If additional environments have been implemented on top of the base code, this may not work. This was only tested on the base code.
To start the installation, navigate to the folder and type:
./install.shAt which point, there is a prompt for the HyperNEAT folder location:
HyperNEAT root location:This should reference the root location of HyperNEAT v4.0.
Provided this was successful, files will be transferred to the correct directories and a patch applied to update existing files. Build instructions can then be followed at Jason Gauci's repo.
Reference the file below to begin the experiment:
HyperNEAT/NE/HyperNEAT/out/data/RiverExperiment.datThe file RiverExperiment.dat contains the parameters used in 'Toward evolving robust, deliberate motion planning with HyperNEAT'. These can be changed to suit preference.