VFSTL

Setup

Pull the project with sub-modules

1. git pull --recurse-submodules [email protected]:harryting-yiting/VFSTL.git

1. Build Docker environemnt

• Prerequisites
  1. docker engine
  2. nvidia-docker-toolkit
• Build Docker image
  1. Change to the project directory
  2. Build the docker with docker build -t harryting/vfstlpy37_piptorch_cuda12.3:v1 .
  3. Run this container by docker run -it --gpus all -v ./:/app/vfstl/src harryting/vfstlpy37_piptorch_cuda12.3:v1

2. Build Conda environment

• 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
  

Experiments

plot

• save the video of specified STL task

python exp.py --type='plot' --stl='eventually[0,5] (J0 >= 0.8)'

run 100 experiment and see the statistics of stl robustness, euclidean distance robustness

• 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