Skip to content

CARLA simulator#

CARLA is a famous open-source simulator for the autonomous driving research. Autoware Universe includes a bridge package for CARLA, autoware_carla_interface. Projects from the community also connect Autoware to CARLA. The document is to list these projects for anyone who wants to run Autoware with Carla. You can report issues to each project if there is any problem.

Project lists (In alphabetical order)#

autoware_carla_interface#

Autoware ROS package that enables communication between Autoware and CARLA simulator for autonomous driving simulation. It is integrated in autoware_universe and actively maintained to stay compatible with the latest Autoware updates.

  • Package Link and Tutorial: autoware_carla_interface.
  • Docker Compose demo: docker/examples/demos/carla in the Autoware repository. It runs CARLA 0.9.16, autoware_carla_interface, and Autoware (e2e_simulator.launch.xml with simulator_type:=carla) in separate containers.

autoware_tensorrt_vad (end-to-end planning)#

TensorRT-optimized Vectorized Autonomous Driving (VAD) node that replaces the traditional perception/localization/planning stack with a single end-to-end model trained on CARLA (Bench2Drive). Ships with a CARLA-focused launch file (vad_carla_tiny.launch.xml) that you start next to e2e_simulator.launch.xml.

  • Project Link: autoware_tensorrt_vad (README includes parameters, topics, and model details)

  • Usage (CARLA E2E mode):

    1. Build the package and deps: colcon build --symlink-install --cmake-args -DCMAKE_BUILD_TYPE=Release --packages-up-to autoware_tensorrt_vad.
    2. Prepare CARLA following autoware_carla_interface (use carla_sensor_kit so camera topics match the VAD training order: FRONT, BACK, FRONT_LEFT, BACK_LEFT, FRONT_RIGHT, BACK_RIGHT).
    3. Make sure that the VAD models are downloaded. The artifacts ansible role downloads them to ~/autoware_data/ml_models/vad/v0.1.
    4. Start CARLA server (example headless/GPU-friendly): ./CarlaUE4.sh -prefernvidia -quality-level=Low -RenderOffScreen.

    5. Launch Autoware in E2E mode:

      ros2 launch autoware_launch e2e_simulator.launch.xml \
  map_path:=$HOME/autoware_data/maps/Town01 \
  vehicle_model:=sample_vehicle \
  sensor_model:=carla_sensor_kit \
  simulator_type:=carla \
  use_e2e_planning:=true

      The use_e2e_planning:=true argument makes autoware_carla_interface publish the ground-truth localization and start a minimal control stack (a trajectory follower and a vehicle command gate).

    6. Start the VAD node in another terminal. The package provides the vad_carla_tiny.launch.xml launch file for it. In Autoware 2.0.0, e2e_simulator.launch.xml does not include this file.

      ros2 launch autoware_tensorrt_vad vad_carla_tiny.launch.xml

      The command in step 5 also starts the localization, perception, planning, and control modules of Autoware. The VAD node and these modules then publish on the same topics, for example /planning/trajectory and /perception/object_recognition/objects. The e2e_simulator.launch.xml file has localization, perception, planning, and control arguments that turn these modules off. Autoware 2.0.0 has no tested launch command for VAD with CARLA that uses them.

carla_autoware_bridge#

An addition package to carla_ros_bridge to connect CARLA simulator to Autoware Universe software.

open_planner#

Integrated open source planner and related tools for autonomous navigation of autonomous vehicle and mobile robots

zenoh_carla_bridge#

The project is mainly for controlling multiple vehicles in Carla. It uses Zenoh to bridge the Autoware and Carla and is able to distinguish different messages for different vehicles. Feel free to ask questions and report issues to autoware_carla_launch.