gym-swirl is a software package for simulating active colloids with social interaction. The simulation is a python implementation of the article "Formation of stable and responsive collective states in suspensions of active colloids" by Bäuerle et al. It is written as an OpenAI Gym package using PyTorch to better suit the needs for a reinforcement learning model.
The software package is tested with python version >=3.8.
Using pip, the package is installed by running the following command in the project root folder:
pip install -e .
To avoid dependency conflicts, using conda or other python environment and package management systems is recommended.
The simulation can easily be run and displayed using a jupyter notebook. For an example usage, see notebooks/simulation.ipynb.
The default arguments, that can be overwritten by passing them into the env.reset() as keyword arguments, are as follows:
defaults = {
"amount": 48, # number of particles
"spread": 5*6.3e-6, # the standard deviation of the initial position in m
"T": 0., # starting time in s
"T_release": 200., # release time in s
"velocity": 0.5e-6, # velocity magnitude in m/s
"diameter": 6.3e-6, # particle diameter in m
"Rc": 6.3e-6/2, # radius of collision in m
"Rr": 8e-6, # radius of repulsion in m
"Ro": 25e-6, # radius of orientation in m
"Ra": float("inf"), # radius of attraction in m
"alpha": 2. * pi, # visibility angle in rad
"DT": 0.014e-12, # translational diffusion constant in m^2 s^−1
"DR": 0.0028, # rotational diffusion constant in s^−1
"Gamma": 25., # max torque factor (unitless)
"dt": 0.2, # time step size in s
"random_angles": True, # boolean for initial random angles
}Author Johan Dettmar 2020-2021.