git pull --recurse-submodules [email protected]:harryting-yiting/VFSTL.git
- Prerequisites
- docker engine
- nvidia-docker-toolkit
- Build Docker image
- Change to the project directory
- Build the docker with
docker build -t harryting/vfstlpy37_piptorch_cuda12.3:v1 . - Run this container by
docker run -it --gpus all -v ./:/app/vfstl/src harryting/vfstlpy37_piptorch_cuda12.3:v1
- Using conda and install
pygraphviz.conda install -c conda-forge pygraphviz
- Install mujoco and mujoco-py
- Install safty-gym.
pip install -e GCRL-LTL/zones/envs/safety/safety-gym/ - Install required packages
conda update --file environment.yaml
- save the video of specified STL task
python exp.py --type='plot' --stl='eventually[0,5] (J0 >= 0.8)'- chain experiments e.g.
$j U (r U (y))$
python exp.py --type='exp' --task='chain'- avoid experiments e.g.
$\neg y U (j \wedge (\neg wUr))$
python exp.py --type='exp' --task='chain'- stable experiments e.g.
$GF(r)$
python exp.py --type='exp' --task='stable'- once the experiment done, user can view result with
controller_evaluation_result/box_plot.ipynb