First Simulation

Run v2xp_datadump_town06_carla in CARLA Town06. Complete Installation first. This example uses CARLA and core OpenCDA. SUMO, Artery, and learned cooperative perception are optional additions in the later guides.

Create the Containers

Run in Linux Bash or Ubuntu in WSL2, from the CAVISE root:

Linux

./run.sh build carla opencda-minimal
./run.sh up carla opencda-minimal

Windows

CARLA runs on Windows, so build only OpenCDA here:

./run.sh build opencda-minimal
./run.sh up opencda-minimal

Check that the container is running and can access the GPU:

docker ps
docker exec opencda nvidia-smi

up creates containers that wait with sleep infinity. Start CARLA and the scenario explicitly in separate terminals below.

Start CARLA

Keep this terminal open while the scenario runs.

Linux

In Linux Bash:

docker exec -it carla bash
./CarlaUE4.sh -quality-level=Low

For off-screen rendering, use ./CarlaUE4.sh -RenderOffScreen instead.

Windows

In Windows PowerShell, change to the extracted CARLA directory containing CarlaUE4.exe, then run:

.\CarlaUE4.exe -quality-level=Low

For off-screen rendering, use .\CarlaUE4.exe -RenderOffScreen instead.

Connect from OpenCDA

In a new Linux/WSL terminal, enter the OpenCDA container:

docker exec -it opencda bash

The shell starts in the mounted OpenCDA repository. Set the CARLA address inside this container shell:

Linux

export CARLA_HOST=carla

Windows

export CARLA_HOST=host.docker.internal

This address is provided by Docker Desktop. A Docker Engine installed independently inside WSL may require a different Windows host address. See Troubleshooting.

All subsequent guides use --carla-host "$CARLA_HOST". Repeat this export whenever you open a new container shell. CARLA_HOST is a shell variable used by these examples. The runner does not read it automatically.

Before launching, query both API versions and the available Town06 map:

python - <<'PYTHON'
import os
import carla
client = carla.Client(os.environ["CARLA_HOST"], 2000)
client.set_timeout(30.0)
print("Client:", client.get_client_version())
print("Server:", client.get_server_version())
print("Town06:", [m for m in client.get_available_maps() if m.endswith("/Town06")])
PYTHON

Both versions should match and Town06 should be listed. Resolve a connection or map error before continuing.

Run the Scenario

In the same OpenCDA container shell:

python opencda.py -t v2xp_datadump_town06_carla \
  --carla-host "$CARLA_HOST" --ticks 200

OpenCDA loads opencda/scenario_testing/config_yaml/v2xp_datadump_town06_carla.yaml over default.yaml, initializes Town06, creates the configured agents, and advances the simulation. --ticks bounds the run. It may end sooner if its scenario completion conditions are met. Logs and evaluation output are under simulation_output/evaluation_outputs/ in the mounted OpenCDA repository.

Use --free-spectator to keep manual control of the CARLA camera. See Recording and Results for recording, visualization, and output files.

Stop and Run Again

Let the bounded run finish, or press Ctrl+C in the OpenCDA terminal. Stop CARLA with Ctrl+C in its terminal when no longer needed. To stop the containers, run from the CAVISE root on Linux/WSL:

Linux

./run.sh stop carla opencda-minimal

Windows

./run.sh stop opencda-minimal

Use run.sh start with the same services to restart existing containers, then launch the simulator processes again. Use up after rebuilding or changing an image target. start does not recreate a container.