This repository provides the code of SoVAR.

The structure of folder "src" is as follows:
src
├── file
│   ├── report                      crash reports                  
│   ├── apiKey.txt                  write your openai api-key here          
│   ├── extractionPrompt.txt        prompt with linguistic patterns            
│   ├── map.txt                     road information  
│   ├── report.txt                  input of SoVAR  
│   └── result.txt                  output of gpt        
├── solver		            solvers of car actions               
├── caculator.py                    main caculator of SoVAR  
├── generateWaypoints.py            generate waypoints          
├── gpt.py                          chat with gpt
├── search.py                       adjust points       
├── simulateReport.py               entrance to simulate report                 
├── testApollo.py                   entrance to test apollo car     
├── testRandom.py                   entrance to test apollo car with random npc
└── utils.py                        some basic methods     

• Python version : 3.10 or higher.

• Basic dependencies : see in requirements.txt.

• Core API : LGSVL PythonAPI.

• Car simulator : LGSVL simulator v2021.3.

• Autopilot platform : Apollo 6.0.

• Operation system : Ubuntu 21.10.

• Hardware condition (Not mandatory) :

  • GPU : Intel (R) Core (TM) i7-1070K.

  • Internal memory : 32GB.

  • Display : NVIDIA GeForce RTX 4070.

1. Install LGSVL

LGSVL simulator can be installed from https://github.com/lgsvl/simulator. The tutorial is on it as well. We are using the latest version,2021.3. LGSVL has made the difficult decision to suspend active development of SVL Simulator, as of January 1, 2022. The cloud had stopped running on June 30, 2022. Therefore, we use SORA-SVL to build our own server as a replacement.SORA-SVL can be installed from https://github.com/YuqiHuai/SORA-SVL

2. Install Apollo6.0

1. clone source code.

$ git clone https://github.com/ApolloAuto/apollo.git

2. pull docker image and enter the container.

$ sudo bash ./docker/scripts/dev_start.sh
$ sudo bash ./docker/scripts/dev_into.sh

3. build Apollo.

sudo ./apollo.sh build

4. start dreamviewer.

sudo bash scripts/bootstrap.sh

5. bridge Apollo with LGSVL.

bash scripts/bridge.sh

3. Install SoVAR

1. clone source code.

2. install dependencies.

pip install -r requirements.txt

3. import LGSVL PythonAPI.

Download python-api from https://github.com/lgsvl/PythonAPI/tree/master. After download, put the package "lgsvl" under the package "src" of SoVAR.

E1: Reconstruct 2-NPCs car accident from an accident report.

• Start lgsvl simulator.
• Copy the summary part of the report in XML format to report.txt.
• Run simulateReport.py.

The two cars' behaviors are the same as the report.

E2: test 1-EGO-and-1-NPC car accident in Apollo from an accident report.

• Start lgsvl simulator.
• Start Apollo6.0 and connect it with lgsvl simulator.
• Copy the summary part of the report in XML format to report.txt.
• Run testApollo.py.

The NPC car's behavior is same as the report while the EGO car has the same start point and destination point as the report.

E3: test 1-EGO-and-1-random-NPC car accident in Apollo from an accident report.

• Start lgsvl simulator.
• Start Apollo6.0 and connect it with lgsvl simulator.
• Copy the summary part of the report in XML format to report.txt.
• Run testApollo.py.

The EGO car has the same start point and destination point as the report. But the random NPC car has a random start point and destination point.

Q : How to run the experiments RQ1, RQ2 and RQ3 in the paper ?

Experiments above are the usage mode of SoVAR in practical application. You can rerun RQ1, RQ2 and RQ3 in the paper by modifying a few codes based on E1, E2 and E3.