oliverweissl / morphologicalnovelty

A method to determine morphological novelty in a set of modular robots.

Thesis

- THIS IS DEPRECATED -> If you want to use Morphological Novelty yourself look at revolve2/ci-group https://github.com/ci-group/revolve2

Prepare for Simulations:

• Install Python 3.10
• Install Revolve2 -> https://ci-group.github.io/revolve2/installation/index.html
• Install Julia 1.8
• in python terminal:
  • pip install julia
• in julia terminal:
  • import Pkg
  • Pkg.add("PyCall")
  • Pkg.add("Shuffle")

Run Simulations:

For one specific simulation:

• python3 -m run_batch --novelty <novelty weigth or None> --amount <amount of simulations> --seed <seed>
  

For running multiple simulations in parrallel:

• sh start_batches.sh

if novelty_weight -> parent selection using weighted average (fitness*(1-novelty_weight) + noveltynovelty_weight)
if None -> parent selection using penalized fitness (fitnessnovelty)

morphologicalnovelty's Issues

pca-cob heuristic approx

pca-cob currently is made for continuous-like distr.
robot body in revolve -> canonly grow on x,y,z axis since quadratic building blocks.

--> exchange of axis instead of cob