CARLA is an open-source autonomous driving simulator based on Unreal Engine 4. The primary ways of interacting with the CARLA setup are through the Python API or with ROS2 (via the CARLA ROS Bridge). Both of these methods are explained in greater detail further down in this document.
The goal of the CARLA setup is to provide an easy way for WATonomous members to interact with the simulator without having to setup anything themselves. So, if you have any questions or suggestions regarding any part of the CARLA setup (or questions about CARLA in general) please bring them up in Discord in #simulation-general or contact Vishal Jayakumar (masteroooogway on Discord or at [email protected]).
Make sure you are confortable with navigating the monorepo before reading
To run CARLA and all associated containers first add simulation as an ACTIVE_MODULE in watod-config.sh. This will cause the following containers to launch when watod up is run: carla_sim, carla_viz, carla_ros_bridge, carla_notebooks, carla_sample_node.
This is the simplest way of interacting with CARLA and most importantly does not require the usage or knowledge of ROS. The documentation below will only show the setup procedure for using the CARLA Python API in the WATOnomous server (via Jupyter Notebook). The full CARLA Python API documentation can be found here (CARLA Python API Documentation).
On the WATOnomous Server
Forward the port specified by the variable CARLA_NOTEBOOKS_PORT (which can be found in the .env file in the modules folder once watod up is run) using either the ports section of VS Code or by running the command ssh -L 8888:localhost:8888 <username>@<machine>-ubuntu1.watocluster.local on your local machine (replace 8888 with the port specified in the CARLA_NOTEBOOKS_PORT variable). If the port is auto-forwarded by VS Code then you do not need to add it manually.
On your local machine (your personal laptop/pc)
Open any web browser (Chrome, Firefox, Edge etc.) and type localhost:8888 (replace 8888 with the port specified in the CARLA_NOTEBOOKS_PORT variable) and click enter. You should be automatically redirected to the Jupyter Notebook home screen where you should see the file carla_notebooks.ipynb. Open that file and follow the instructions in it. Either edit the carla_notebooks.ipynb file directly or create a new file and copy the first code block in carla_notebooks.ipynb.
Ensure that you have a good understanding of ROS and writing ROS nodes before proceeding
Currently there is a CARLA sample node setup (written in Python) shows how to publish a message to enable (and keep enabled) autopilot, subsribe to the GNSS sensor topic and output the data to the console. What is probably most helpful from the sample node is the Dockerfile located in wato_monorepo_v2/docker/simulation/carla_sample_node/carla_sample_node.Dockerfile, where you can find all the CARLA related messages that should be installed when building your own ROS nodes that interact with CARLA.
Foxglove Studio is a tool to visualize and interact with ROS messages. Using Foxglove Studio, data such as Camera, LiDAR, Radar etc. can be visualized. Foxglove Studio also allows for messages to be published to topics, for example to enable autopilot or set vehicle speed. The documentation below will only show the setup procedure for using Foxglove Studio with the WATOnomous server. Further documentation regarding how to use Foxglove Studio and all its features can be found here (Foxglove Studio Documentation)
On the WATOnomous Server
Add infrastructure as an ACTIVE_PROFILE. This will launch the foxglove container with an open port when watod up is ran.
It exposes the port specified by the FOXGLOVE_BRIDGE_PORT variable, which will be defined in the .env file in the modules folder (the .env file is populated after watod up is run). This port may be auto-forwarded by VS Code automatically but if not the port will need to be forwarded manually to your local machine. This can either be done in the ports section of VS Code or by running the command ssh -L 8765:localhost:8765 <username>@<machine>-ubuntu1.watocluster.local on your local machine (replace 8765 with the port specified in the FOXGLOVE_BRIDGE_PORT variable).
On your local machine (your personal laptop/pc)
Open Foxglove Studio on your local machine using either the desktop app (Download Here) or the Web Application (only supported on Chrome). Click on Open Connection then replace the default port (8765) with the port you specified in the FOXGLOVE_BRIDGE_PORT variable. Then click the Open button and after a few seconds Foxglove Studio should connect. You should now be able to access any of the topic being published or subscribed by the CARLA ROS Bridge.
On the WATOnomous Server
CarlaViz is a visualization tool that is useful when you are only using the CARLA Python API through a Jupyter Notebook and you don't want the overhead that comes with the ROS Bridge + Foxglove method of visualization. To use CarlaViz forward the ports specified by the variables CARLAVIZ_PORT and CARLAVIZ_PORT_2 defined in the .env file in the modules folder (make sure you forward both ports). If you want to stop the ROS Bridge from running (for the reasons I mentioned in the previous sentences) then make sure to comment out the carla_ros_bridge service in the docker-compose.carla.yaml file in the modules folder.
On your local machine (your personal laptop/pc)
In any browser go to localhost:8081 (replace 8081 with the port specified in the CARLAVIZ_PORT variable) and after waiting 10 seconds or so CarlaViz should appear.
This is expected. The ROS bridge causes CARLA to render all the sensor data which slows down the simulation considerably. While this may be annoying when viewing real-time camera output or trying to control the car manual, the simulation is still running accurately. If you wish to replay scenes with a configurable playback speed, check out the docs on CARLA's recorder functionality.