Skip to content

Latest commit

 

History

History
48 lines (27 loc) · 2.26 KB

INSTALL.MD

File metadata and controls

48 lines (27 loc) · 2.26 KB
Raw

INSTALL

The following guide serves as a user manual for running DoppelTest to realize Doppelganer Testing on a production-grade ADS (Baidu Apollo). The installation is divided into 2 parts - 1) installing the ADS and 2) installing DoppelTest.

INSTALLING Baidu Apollo

  1. Download the DoppelTest version of Apollo 7.0 from https://doi.org/10.5281/zenodo.7622089

In this forked version, we made slight adjustments that are not related to the AD stack

  1. At the root directory of Baidu Apollo, create directories data, data/log, data/bag, data/core by running mkdir data data/log data/bag data/core

This step is necessary for DoppelTest running on the host machine to delete Apollo's log files. Our framework restarts modules being tested after every scenario, which creates a large number of unnecessary log files.

Since a lot of commands are executed as root inside of the Docker container, if those directories are created inside of the container, DoppelTest may not be able to remove those directories.

  1. At the root directory of Baidu Apollo, start up Apollo container via ./docker/scripts/dev_start.sh -l

  2. Find the name of the container via docker ps -a

  3. Enter the container in root mode via docker exec -it apollo_dev_your_name /bin/bash

Remember to replace apollo_dev_your_name with the container's actual name

  1. In the container, build Apollo via ./apollo.sh build

INSTALLING DoppelTest

  1. Install the required Python libraries via pip install -r requirements.txt

If you run into issues when installing Shapely library, please first run sudo apt-get install libgeos-dev to install its dependencies.

  1. Replace location of directories in config.py

    APOLLO_ROOT  = '/xxx/xxx/apollo'
DT_ROOT      = '/xxx/xxx/DoppelTest'

  2. Verify the framework is runnable via python test_main.py

You should start seeing 3 Apollo instances being started and the scenario is visualizable via a browser. DoppelTest will provide the URL to visualize each instance in the terminal.

  1. Start the framework via python main_ga.py

After running DoppelTest for extended period of time, you should see record file of scenarios generated under data/records. This is also the step to replicate the results presented in the paper.