carla_interface

Provides an interface back and forth between Navigator messages and the CARLA simulator.

• At its core is the ros_carla_bridge which sends sensing data from CARLA to a ROS autonomy stack. It also receives actuator commands from the ROS stack and enacts them in CARLA on the ego vehicle. The ros_carla_bridge repository supports ROS2 Foxy (not Humble), which is partially why we have a dedicated Docker container for it.
• The ros_carla_bridge also loads the map, configures the basic scene, and configures the ego vehicle (including which sensors it has).
• Beyond renaming some of the topics published directly by the ros_carla_bridge (this occurs in the launch file script), this repository also provides a few nodes to process sensing data from CARLA before presenting it to the autonomy stack. The goal of these nodes is to publish the following topics for the autonomy stack:
  • /planning/rough_route (published by carla_interface/route_reader.py)
  • /gnss (published by carla_interface/carla_gnss_processing_node.py)
  • /lidar (published by carla_interface/carla_lidar_processing_node.py)
  • /imu (renamed in launch file)
  • /cameras/camera0 (front camera, renamed in launch file)
  • /cameras/camera1 (right camera, renamed in launch file)
  • /cameras/camera2 (back camera, renamed in launch file)
  • /cameras/camera3 (left camera, renamed in launch file)
  • /depth/camera0 (front camera, depth, renamed in launch file)

Usage:

1. Run the docker image: docker compose run carla_bridge
2. Launch: ros2 launch launch.carla_interface.py

Configuring the Ego Vehicle

The ego vehicle is configured within carla_interface/data/ego_config.json. This file is read by the launch file. The vehicle type, spawn point, and other features of the vehicle are set in this file. Sensors along with their locations and orientations are specified here. There is a property called autopilot that can be made true or false to make the vehicle drive on its own.

The ego vehicle can be manually driven in CARLA by uncommenting the carla_manual_control node and commenting out the carla_vehicle_control node. This opens a dialog box that allows you to use the keyboard to navigate the vehicle when that dialog box has focus. This dialog also has an option to enable autopilot there, which is another way to make the vehicle drive on its own.

Useful for recording videos of the vehicle, there is a /carla/hero/rgb_overhead/image topic that is published by a camera floating above and to the left of the vehicle.

Spawning Vehicles and Pedestrians

We have several ways to spawn other actors in the envrionment.

• A ROS node vehicle_spawner will spawn vehicles
• The CARLA PythonAPI has an example code to spawn a given number of vehicles and pedestrians randomly. The carla directory is mounted at / within the docker container.

python3 /carla/PythonAPI/examples/generate_traffic.py --port 20XX -n 100 -w 50 --safe where the port should match 2000 plus your ROS_DOMAIN_ID (set in your environment variables), n is the number of vehicles, and w is the number of walkers. The --safe flag is supposed to spawn vehicles that drive safely (this doesn't seem to work perfectly).