neural-cellular-robot-substrate

Neural cellular automaton for developing body and brain of a modular robot

Code and videos for the paper: arXiv Preprint

Dependencies used:

• Python 3.8.10
• TensorFlow 2.6
• gym 0.21.0
• pygame 2.0.1
• box2d-py 2.3.5
• cma 3.1.0
• numpy 1.19.5
• moviepy 3.1.1
• imageio 2.9.0

Folder: src

Playing light chasing task:

python modular_light_chaser.py

Playing carrying ball to target:

python modular_carrier.py

Train light chasing task with CMA-ES:

python train_modular_light_chaser_es.py

Train light chasing with obstacle task with CMA-ES:

python train_modular_light_chaser_es.py --wall

Train carrying ball to target task with CMA-ES:

python train_modular_carrier_es.py

Display training options using python train_modular_light_chaser_es.py --help:

usage: train_modular_light_chaser_es.py [-h] [--maxgen MAXGEN] [--popsize POPSIZE] [--threads THREADS] [--width WIDTH] [--height HEIGHT] [--stdinit STDINIT] [--bodytype BODYTYPE] [--retrain RETRAIN] [--saveall] [--wall]

optional arguments:
  -h, --help           show this help message and exit
  --maxgen MAXGEN      Maximum number of generations
  --popsize POPSIZE    Population size
  --threads THREADS    Number of threads
  --width WIDTH        CA width
  --height HEIGHT      CA height
  --stdinit STDINIT    Standard deviation to initialize the Evolution Strategy
  --bodytype BODYTYPE  Premade body types: {1,2,3}
  --retrain RETRAIN    path to log directory for retraining
  --saveall            Save all solutions
  --wall               Wall environment

Test trained NCA:

python test_modular_*.py --logdir <path>

Generate test video of trained NCA:

python test_video_modular_*.py --logdir <path>

Folder: videos

This folder contains sample videos of the trained models reported in the paper.

Citation

@inproceedings{pontes2022unified,
  title={A Unified Substrate for Body-Brain Co-evolution},
  author={Pontes-Filho, Sidney and Walker, Kathryn and Najarro, Elias and Nichele, Stefano and Risi, Sebastian},
  booktitle={From Cells to Societies: Collective Learning across Scales},
  year={2022}
}