visualize-ros-rviz

Inspects and visualises a ROS-2-enabled CARLA server from Docker containers, with no local ROS 2 install — list/echo/rate-check topics (the "are the topics actually on the wire" check), run the bundle

SKILL.md

GitHub

See the ROS 2 side of a CARLA server

Paths. scripts/… and references/… below are relative to the directory holding this SKILL.md. Your working directory is the user's project, not that directory, so prefix them with its absolute path or the command is not found.

Everything else in this collection acts through the RPC API. This skill is the consumer side: it proves what is really on the DDS wire and draws it. It never launches a simulator, and it needs no ROS 2 on the host — the ros2 CLI and RViz run inside images built from the checkout's own PythonAPI/examples/ros2/.

Use it after [[run-carla-server]] ROS2=1: world-data ros-topics says what should be published, this says what is.

Instructions

Progress:
- [ ] Step 1: Check prerequisites (bash scripts/check_env.sh), clear FAILs
- [ ] Step 2: A server is running WITH ROS 2, on a known --rmw and domain
- [ ] Step 3: topics -> prove /clock exists; echo/hz -> prove data flows
- [ ] Step 4: (opt-in) demo -> spawn the stack; rviz -> draw it

Step 1: Check prerequisites

bash scripts/check_env.sh

Step 3: Inspect (read-only)

source scripts/env.sh

bash scripts/ros_view.sh topics                      # ros2 topic list + verdict
bash scripts/ros_view.sh info /carla/map             # type, endpoints, QoS
bash scripts/ros_view.sh echo /carla/map             # one sample (latched OpenDRIVE)
bash scripts/ros_view.sh hz /carla/hero/lidar/point_cloud 10
bash scripts/ros_view.sh local-env                   # exports for a LOCAL ros2 install

echo adds --qos-durability transient_local --full-length automatically for /carla/map: it is latched and published only on episode start, so a default (volatile) subscription waits forever and a default echo truncates the OpenDRIVE at 128 characters.

Using a local ROS 2 instead of these containers? Run eval "$(bash scripts/ros_view.sh local-env)" first. It exports the checkout's RMW profile, which forces UDP-only transport. Without it CARLA's shared-memory-built Fast DDS never matches your subscriber: the topics list fine and no data ever arrives, with no error on either side.

topics is the one to run first: it waits for DDS discovery, lists the topics and fails loudly with the ordered list of causes when /clock is absent (not built with ROS 2 → not started with --ros2 → domain mismatch → RMW mismatch → missing Zenoh router).

Pass ROS names (/clock), not DDS names (rt/clock) — the rt/ prefix is the DDS-level encoding of the same topic.

Step 4: Demo stack and RViz (opt-in — has side effects)

bash scripts/ros_view.sh demo            # SPAWNS a hero + camera/lidar/GNSS/IMU
bash scripts/ros_view.sh demo --map-only # only the lane markers, no actors
bash scripts/ros_view.sh rviz            # RViz2 with the bundled preset

demo mutates the world: it spawns a hero vehicle with four sensors, drives it on autopilot, publishes map->hero TF and converts the latched /carla/map OpenDRIVE into /carla/map_markers for RViz. Run it only when the request asks for the demo or for a live scene to visualise; it cleans up its actors and restores the world settings on exit (single Ctrl+C). A second concurrent run fails fast on the fixed container name instead of spawning a duplicate hero on the same topics.

rviz is read-only and needs $DISPLAY. Both modes delegate to the checkout's own run_map_and_lidar_demo.sh / run_rviz.sh, so their behaviour is never forked here.

On CARLA 0.10.0 (the UE5 line: 5.5 and 5.8)

/carla/map is gone on 0.10.0. CarlaMapPublisher exists in 0.9.x's LibCarla/source/carla/ros2/publishers/ and is absent in 0.10.0, so the latched OpenDRIVE topic is never advertised. info /carla/map and echo /carla/map will simply wait. Get the map with map.to_opendrive() over RPC, or point RViz at the lidar/camera topics, which are unchanged.

That also means the RViz lane-network preset has no map source on 0.10.0 — the road network has to come from the client side.

Topics that gained publishers on 0.10.0, UE 5.8 only: the Autoware set (AutowareGNSSPublisher, AutowareVehicleStatusPublisher, from sensor.other.autoware_gnss and sensor.other.vehicle_status). The DVS camera publisher was renamed CarlaDVSPublisher → CarlaDVSCameraPublisher; its topics are unchanged.

rt/tf is additionally gated by World.set_publish_tf() on 0.10.0, so a missing TF tree can now mean "switched off globally" rather than "no sensor publishing".

Knobs

Env Default Meaning
ROS_DISTRO_TAG humble humble or jazzy; picks the image and the wheel tag (cp310 / cp312)
RMW fastdds fastdds · cyclonedds · zenoh — must match the server's --rmw
ROS_DOMAIN_ID unset (0) must match the server's --ros-domain-id
CARLA_HOST/CARLA_PORT localhost/2000 the server the demo stack connects to

Images are named carla-rviz-<distro>-<rmw> and carla-map-and-lidar-demo-<distro>-<rmw>, so each combination is built once and cached. Only the demo image needs the carla wheel; topic inspection and RViz do not.

Examples

Example 1: is CARLA publishing at all?

User says: "is CARLA publishing to ROS?"

bash scripts/ros_view.sh topics. /clock present → yes, and the list shows which actor topics exist. Absent → the command prints the ordered causes.

Example 2: the camera topic is missing

User says: "my camera doesn't show up in ROS"

topics (server alive?) → [[world-data]] ros-topics (does the sensor exist and is it enabled_for_ros?) → [[create-sensor]] ros --id N to enable it → hz on the topic to confirm frames.

Example 3: see the town and the lidar together

User says: "show me the lidar in RViz"

demo in one shell (spawns the stack), rviz in another. Fixed frame map; set Views → Target Frame: hero to follow the car.

Verify

  • topics exits non-zero when /clock is missing — treat that exit code as the gate, not the printed list.
  • info <topic> shows the publisher count: 0 means the topic name exists in discovery but nothing writes it.
  • hz gives the real rate; a camera at ~0 Hz with a live server usually means -nullrhi (no rendering, so no images) rather than a ROS problem.

Troubleshooting

Problem: topics lists nothing at all (not even /clock) Cause: in order of likelihood — the server was not built with ROS 2, was started without --ros2, is on a different ROS_DOMAIN_ID, or uses a different --rmw. Solution: work down that list; the command prints it. Note RMW does not have to equal the server's --rmw for fastdds/cyclonedds — verified interoperable, both being RTPS. Only zenoh must match on both sides.

Problem: topics appear but echo/hz gets nothing (local ROS 2) Cause: no RMW profile exported — shared-memory vs UDP mismatch. The most common local failure, and completely silent. Solution: eval "$(bash scripts/ros_view.sh local-env)". The containers here are unaffected.

Problem: topics appear but echo hangs (in the containers too) Cause: subscribing to a best-effort sensor topic that is not currently producing (sensor not enable_for_ros-ed, or a camera on a -nullrhi server); or a latched topic subscribed volatile. Solution: [[world-data]] ros-topics to see the enable state; use PACKAGED=1 or WINDOW=1 for camera data ([[run-carla-server]]).

Problem: a topic has a publisher but never any data, after a map change Cause: CARLA does not unregister ROS 2 publishers on episode teardown, so the old actor's endpoint lingers (verified: Publisher count grows by one per map switch when the same ros_name is reused). Solution: ignore the zombie or restart the server; see [[load-map]].

Problem: RViz opens empty / no map markers Cause: RViz started before the latched sample was published, or the demo stack is not running (the markers come from the demo's map_to_markers.py, not the server). Solution: toggle the Map display checkbox to resubscribe, and make sure demo is running (--map-only is enough for markers alone).

Problem: docker: permission denied / daemon unreachable Cause: the user is not in the docker group. Solution: fix the group membership outside this skill; every mode here needs the daemon.

Problem: zenoh: server and containers see nothing Cause: no Zenoh router. Zenoh is the only RMW here needing a separate process. Solution: start rmw_zenohd before both, e.g. docker run --rm --net=host carla-rviz-humble-zenoh ros2 run rmw_zenoh_cpp rmw_zenohd.

Outputs

Console output (topic lists, samples, rates) and, for rviz, a window. demo leaves the world as it found it. Nothing is written to the checkout.

Consumer-side detail — QoS matching, use_sim_time, the demo's helper processes, the topic-name mapping — in references/ros-consumers.md.