Runtime Commands and Targets

Use First Simulation for a complete first run. This page describes command choices and advanced runtime operations.

Repository Setup

Run python setup.py in the CAVISE root, with its Python environment active. Omitting repository names selects all five repositories. Omitting a version opens an interactive branch/tag selection. It does not silently select main. Existing directories are skipped.

Repository	Version option
`opencda`	`-o` / `--opencda-version`
`opencood`	`-O` / `--opencood-version`
`sumo`	`-S` / `--sumo-version`
`artery`	`-a` / `--artery-version`
`scenario-manager`	`-s` / `--scenario-manager-version`

For example, clone OpenCDA and OpenCOOD at explicit refs:

python setup.py opencda opencood -o main -O main

Container Lifecycle

Run ./run.sh COMMAND [SERVICES...] from the CAVISE root on Linux/WSL. Main services are carla, opencda, artery and sumo. Omitting services selects all main services, including CARLA on Windows. Prefer an explicit service list for the experiment you are running.

Command	Behavior
`build`	Build images. Does not start containers or simulator processes.
`up`	Create/start containers in the background, recreating them when needed.
`start`	Start existing stopped containers. Does not apply image/config changes.
`stop`	Stop containers while retaining them.
`restart`	Restart existing containers. Does not rebuild images.
`down`	Remove containers via Compose. Files outside bind mounts may be lost.

For example:

./run.sh build opencda-coperception
./run.sh up opencda-coperception
./run.sh stop opencda-coperception
./run.sh start opencda-coperception

Only one OpenCDA target can be selected per invocation. All targets use the same service and container name, opencda. They are not simultaneous independent containers.

OpenCDA Image Targets

Target	OpenCOOD	Protobuf	Custom CUDA	Use
`opencda-minimal`	No	No	No	Core CARLA/OpenCDA, optionally SUMO.
`opencda-protobuf`	No	Yes	No	Artery without cooperative perception.
`opencda-coperception`	Yes	No	No	Models that do not require custom CUDA extensions.
`opencda-cuda`	Yes	No	Yes	Models requiring custom CUDA extensions, such as FPV-RCNN.
`opencda`	Yes	Yes	Yes	Cooperative perception together with Artery. Default target.

All targets use the CUDA runtime base. CUDA extensions are a separate build capability, not the switch that enables GPU computation. opencda-minimal and opencda-protobuf do not install OpenCOOD.

The default CUDA architecture is 86. Set CUDA_ARCHITECTURES to the architectures supported by your GPU and the image’s CUDA compiler, for example:

CUDA_ARCHITECTURES="86;89" ./run.sh build opencda-cuda

Rebuild after changing a Dockerfile or dependencies. Rebuild a Protobuf target after changing a CAPI .proto file, and a CUDA target after changing native OpenCOOD sources. Use up afterwards: native artifacts are synchronized into the mounted workspace by the container entrypoint.

Direct Compose Access

run.sh normally selects the target, tag, paths and model directory for you. For direct Compose use, stay in the CAVISE root and prepare the model mount before creating a container:

source paths.conf
mkdir -p "$PATH_TO_MODELS"
export OPENCDA_BUILD_TARGET=opencda-protobuf
export OPENCDA_IMAGE_TAG=protobuf
docker compose -f dc-configs/docker-compose.yml --env-file paths.conf build opencda
docker compose -f dc-configs/docker-compose.yml --env-file paths.conf up -d opencda

Target suffixes map to image tags minimal, protobuf, coperception and cuda. The full target uses local. Set both variables consistently. To inspect the rendered configuration without starting anything:

docker compose -f dc-configs/docker-compose.yml --env-file paths.conf config

To remove the entire main Compose project:

./run.sh down

Scenario Manager

The web scenario manager uses a separate Compose project and is not included in an unqualified run.sh up. Run it separately:

./run.sh build scenario-manager
./run.sh up scenario-manager

Open http://localhost if the browser does not open automatically. Stop it with ./run.sh stop scenario-manager. Keep these invocations separate from main simulator services because the wrapper forwards service arguments to the selected Compose projects.

OpenCDA CLI

Inside the OpenCDA container, run:

python opencda.py --help

-t selects a YAML path below opencda/scenario_testing/config_yaml without its suffix. Nested paths are accepted. --carla-host defaults to carla. Docker Desktop on Windows uses host.docker.internal in these guides. --artery-host defaults to artery:7777.

Use SUMO and Artery, Cooperative Perception, AdvCP and Recording and Results for complete launch examples.

Artery Frontends

In the Artery container, from /workspaces/artery, select a Qt frontend:

./tools/run_artery.py -l /cached-build/Debug/run-artery.ini -s scenarios/capi -u Qtenv

Use -u Cmdenv for a terminal frontend. Add -c CONFIG_NAME to select a configuration declared by the scenario’s omnetpp.ini. Build instructions and the full startup order are in SUMO and Artery.