This is the official repository for the ALIFE'22 (Artificial Life conference, 2022) paper and its Artificial Life journal extension

Pressure-based Soft Agents
Federico Pigozzi

Shape Change and Control of Pressure-based Soft Agents
Federico Pigozzi

hosting all the code for replication. More videos available at this link.

Clone the repo:

git clone https://github.com/pigozzif/PressureSoftAgents.git

Either install Python dependencies with conda:

conda env create -f environment.yml
conda activate pybox2d

or with pip:

pip install -r requirements.txt

By running

python main.py

you will launch an evolutionary optimization for the controller (an artificial neural network) of Pressure-based Soft Agents (PSAs): they are bodies of gas enveloped by a chain of springs and masses, with pressure pushing on the masses from inside the body. Pressure endows the agents with structure, while springs and masses simulate softness and allow the agents to assume an infinite gamut of shapes. Actuation takes place by changing the length of springs or modulating global pressure. At the same time, evolution metadata will be saved inside the output folder.

Inside config.yaml you may edit the following parameters:

where {...} denotes a finite and discrete set of possible choices for the corresponding argument. The description for each argument is as follows:

• n_masses: the number of rigid masses in the envelope.
• r: the radius of the agent.
• size: label for the size of the agent (just for naming the logs dir).
• solver: the evolutionary algorithm to perform optimization with.
• task: the task to experiment with.
• evaluations: the total number of fitness evaluations before stopping evolution.
• mode: random stands for a random controller, best loads the .npy file for the corresponding experiment, opt-parallel is full-fledged evolution from scratch.
• seed: the random seed.
• np: the number of processes to perform evolution with. Parallelization is taken care by the code and implements a distributed fitness assessment.
• control_pressure: if 1, control also pressure, otherwise just the springs length.
• save_video: if 1, saves simulation to video.mp4.

Please cite as:

@article{pigozzi2023pressure,
  title={Pressure-based soft agents},
  author={Pigozzi, Federico},
  journal={Artificial life},
  pages={1--19},
  year={2023},
  publisher={MIT Press One Broadway, 12th Floor, Cambridge, Massachusetts 02142, USA~…}
}

@proceedings{pigozzi2022shape
    author = {Pigozzi, Federico},
    title = "{Shape Change and Control of Pressure-based Soft Agents}",
    volume = {ALIFE 2022: The 2022 Conference on Artificial Life},
    year = {2022},
    doi = {10.1162/isal_a_00520}
}