Regenerating soft robots through neural cellular automata

Regeneration

Biped

Tripod

Multiped

How to run

First, install Anaconda as python 2.7 distribution on your linux machine.

# clone project   
git clone https://github.com/KazuyaHoribe/RegeneratingSoftRobots.git   

# install dependencies    
cd RegeneratingSoftRobots 
pip install -r requirements.txt

You can run the genetic algorithm using the following command:

python main_creatures.py --number_neighbors 7 --popsize 50 --generations 101 --sigma 0.03 --N 10 --threads 1 --fig_output_rate 10

To build the physical simulator "VoxCad" follow these directions: https://github.com/skriegman/evosoro

After evolving soft robots, you can test their regeneration using the following command:

python regeneration_task.py --number_neighbors 7 --popsize 50 --generations 101 --sigma 0.03 --N 10 --threads 1 --fig_output_rate 10

All arguments can be shown with the help command.

python main_creatures.py

--im_size N           Size of creature
--voxel_types N       How many different types of voxels
--number_neighbors N  Number of neighbors
--initial_noise N     initial_noise
--sigma SIGMA         Sigma
--N N                 N
--simtime SIMTIME     Simulation time in voxelyze
--initime INITIME     Intiation time of simulation in voxelyze
--fraction FRACTION   Fraction of the optimal integration step. The lower, the more stable (and slower) the simulation.
--run_directory RUN_DIRECTORY
--run_name RUN_NAME
--popsize POPSIZE     Population size.
--generations GENERATIONS Generations.
--threads N           threads
--optimizer N         ga
--recurrent N         0 = not recurrent, 1 = recurrent
--data_read N         0 = not data read, 1 = dataread
--cell_sleep N        0 = not sleep, 0 = sleep
--growth_facter N     0 = not growt facter, 1 = growth facter exist
--seed N              seed
--folder N            folder to store results
--show N              visualize genome
--expression N        seperate gene expression
--fig_output_rate N   fig_output_rate

Reproduction of paper's results

We used these parameters for each experiments in both the recurrent and non-recurrent setup. Task _ID means the number of independent running.

2D soft robots

codes: main_creatures_2d, evaluation_2d, GA_2d, read_write_voxelyze_2d
--popsize 300
--generations 501
--sigma 0.03
--N 60
--recurrent 0/1 --seed TASK_ID

3D soft robots

codes: main_creatures, evaluation, GA, read_write_voxelyze
--dimension 1
--initial_noise 1
--popsize 100
--N 20
--generations 301
--sigma 0.03
--recurrent 0/1
--seed TASK_ID

Regeneration task

codes: regeneration_task
--popsize 1000
--generations 1001
--sigma 0.03
--N 200
--recurrent 0/1
--seed TASK_ID

Regeneration through differential calculation(Update 2022/01/06)

If using this part of code please consider citing the following paper.

Sudhakaran, Shyam, et al. "Growing 3D Artefacts and Functional Machines with Neural Cellular Automata." arXiv preprint arXiv:2103.08737 (2021).

You can reproduce the results of the paper by running notebook/CustomVoxel.ipynb.