The env uses OpenSIm RL. It includes an ocular environment and an agent trained using Deep Deterministic Policy Gradients method to perform saccades. The agent was able to match the desired eye position with a mean deviation angle of 3.5°±1.25°.
The proposed DRL environment is based on OpenAI (Brockman et al., 2016), OpenSim (Kidziński et al., 2018a; Seth et al., 2018) and ocular biomechanics (Iskander et al., 2018d, 2019, 2018b).
Watch the RL agent learning to perform saccades
conda create -n opensim-rl -c kidzik -c conda-forge opensim python=3.6.1
source activate opensim-rl
pip install osim-rl
conda install -c anaconda scipy
conda install pytorch torchvision cudatoolkit=10.2 -c pytorch
git clone https://github.com/jIskCoder/jEyeRL.git
source activate opensim-rl
cd jEyeRL
python jeye.py
If you use the code or data in this package, please cite:
@article{ISKANDER2022110943,
title = {An ocular biomechanics environment for reinforcement learning},
journal = {Journal of Biomechanics},
volume = {133},
pages = {110943},
year = {2022},
issn = {0021-9290},
doi = {https://doi.org/10.1016/j.jbiomech.2022.110943},
url = {https://www.sciencedirect.com/science/article/pii/S0021929022000021},
author = {Julie Iskander and Mohammed Hossny},
}
@article{iskander2018ocular,
title={An ocular biomechanic model for dynamic simulation of different eye movements},
author={Iskander, J and Hossny, Mohammed and Nahavandi, Saeid and Del Porto, L},
journal={Journal of biomechanics},
volume={71},
pages={208--216},
year={2018},
publisher={Elsevier}
}