Prerequisites

1. A computer powerful enogh for running Apollo+LGSVL-2021.1. (Or two computers: one for Apollo, one for LGSVL)
2. rtamt (evaluating robustness degree of an STL formula, please refer to the github page, downgrade antlr4 under 4.8)
3. python3

   sudo pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113
   sudo pip3 install websockets pandas sklearn tqdm wsaccel shapely

4. Downgrade protobuf to version 3.19.

Step by step

Run Apollo with LGSVL

Please refer to the detailed documentation for co-simulation of Apollo with LGSVL. Set the LGSVL to API-Only mode.

Copy the content in map/san_francisco folder to Apollo: apollo/modules/map/data/san_francisco/

Setup our bridge.

1. Download and go to the root. Note that the source code should be downloaded and set up on the computer running Apollo.

   git clone https://github.com/researcherzxd/ABLE.git
   cd ABLE-SourceCode

2. Install Python API to support for LGSVL-2021.1.

   cd ABLE-SourceCode/bridge/PythonAPImaster
   pip3 install --user -e .  
   ##If "pip3 install --user -e ." fail, try the following command:
   python3 -m pip install -r requirements.txt --user .

3. Connect our bridge to the LGSVL and Apollo: Go the bridge in the folder:/ABLE-SourceCode/bridge

   cd /ABLE/bridge

   Find file: bridge.py. There is class Server in bridge.py.

   Modify the SIMULATOR_HOST and SIMULATOR_PORT of Server to your IP and port of LGSVL. Modify the BRIDGE_HOST and BRIDGE_PORT of Server to your IP and port of Apollo.

4. Test the parser: If the support for parser is properly installed, we can test it by running:

   cd /ABLE
   python3 monitor.py

   If there is no errors and warnings, the parser is correct.

Run our bridge.

Open a terminal on the computer running Apollo.

cd /ABLE/bridge
python3 bridge.py

Keep it Running.

Run the Generation Algorithm.

Open another terminal on the computer running Apollo.

cd /ABLE
python3 GFN_Fuzzing.py

If the brige is set properly, you will see the LGSVL and Apollo running. The results will be put into a folder that you set in path_config.py.

Docker version

We have wrapped ABLE in Dockerfile.

If you can run Apollo in docker, then you're already good to go.

The following steps are for setting up the docker image. You need to run only once.

0. update the path in testing_engines/gflownet/path_config.py to the following data:

"test_result_direct": "/tmp/ABLE_output/data/apollo7/active+max/{}",
"debug_result_direct": "/tmp/ABLE_output/data/apollo7/debug/{}",

1. create an output folder at path output_dir, say: /tmp/output_dir

2. update the output directory path in the docker_start.sh file

3. open a terminal and bash docker_build.sh

The following steps are for running ABLE:

1. open a terminal and run:
   bash docker_start.sh.
   This will run a docker container with all the dependencies installed.

2. within the docker container of step 1, run bridge:
   cd bridge && python bridge.py

3. open another terminal and run:
   bash docker_enter.sh.
   This will enter the same docker container in step 1, run fuzzing using:
   cd testing_engines/gflownet && python GFN_Fuzzing.py