Rviz2

rviz2

The local-desktop ROS 2 visualization tool: displays for topics, TF frames, markers, and robot models, driven by a saved .rviz config file per robot/task. RViz2 ships across every ROS 2 distro back to the earliest releases through Lyrical Luth (current LTS) and Rolling — verified via direct fetch of index.ros.org's rviz2 package page on 2026-07-10 (the docs.ros.org page returned an access-control error rather than content; see References) — so nothing in this skill is distro-gated the way Nav2's binary packages are; the only distro requirement is that ROS_DISTRO matches the sourced workspace. It is a standard Qt/OpenGL desktop app with no headless or web mode; for a remote/headless server or sharing a view, use foxglove instead (see Platform gotchas).

When to use this skill

• Adding or configuring a display (LaserScan, PointCloud2, RobotModel, TF, MarkerArray, Path, Costmap, Odometry, ...), debugging why a topic isn't rendering, or setting up a saved config for a robot/task.
• The trigger phrases in the description: 'rviz', 'rviz2', visualizing ROS topics, TF trees, costmaps, or robot models during development.
• A display is blank/nothing shows up — start here with the checklist in Usage patterns before assuming the publishing node itself is broken.
• Cross-references — go to the sibling skill instead when the question is:
  • TF2 concepts themselves (frames, broadcasters, static vs dynamic transforms) or QoS concepts (reliability/durability policies, why a mismatch drops messages) → ros2. This skill only applies those concepts as debugging steps (see the nothing-shows-up checklist); it does not re-teach them.
  • Nav2-specific behavior-tree/costmap internals (what a costmap layer computes, BT Navigator internals) → nav2. This skill only lists which displays to add for Nav2 debugging, not what the underlying servers do.
  • Remote/headless viewing, sharing a view with someone not at the machine, or recording for later → foxglove.
  • ML/data-centric logging (policy rollouts, non-ROS sensor pipelines) → rerun.
  • Choosing which viz tool fits the situation at all → visualization (routes here once RViz2 is the right choice).

Key directives

• Delegation posture: embed. RViz2's own domain (displays, config files, the nothing-shows-up checklist) is small and stable enough to live entirely in this skill — there is no large reference tree to split out, and no upstream single page covers the debugging checklist as a connected whole.
• Diagnose in order: Fixed Frame, then QoS, then sim time. These three cover the overwhelming majority of "I added a display and nothing rendered" reports, in that order of likelihood — see the checklist in Usage patterns. Don't jump to restarting nodes or rebuilding the workspace before walking it.
• Never re-teach TF2 or QoS from scratch here. When the checklist points at a broken TF chain or a QoS mismatch, apply the fix (set Fixed Frame, flip the display's QoS override) and link to ros2's TF2/QoS references for why — duplicating that explanation here would drift out of sync.
• Config files are the unit of reuse, not manual per-session setup. Save a named .rviz config per robot/task once the display set is right, and launch with it (-d) every time rather than rebuilding the display list by hand — see Quick start.

Quick start

1. Launch RViz2 standalone:

ros2 run rviz2 rviz2

2. Add a display via the "Add" button (bottom-left panel) — by topic (browses active topics and picks the matching display type) or by display type (e.g. RobotModel, TF, LaserScan) and set its Topic manually.

3. Set the Fixed Frame (Global Options panel, top of the Displays panel) to a frame that actually exists and is being broadcast — usually map or odom for a mobile robot, base_link for a static/manipulator view. This is the single most common cause of a display staying empty; see Usage patterns.

4. Save the config once the display set is right: File > Save Config As..., into a path you'll reuse (e.g. config/<robot>.rviz in the project). Reload it next time with -d (see Usage patterns) instead of rebuilding displays by hand.

Usage patterns

Launch with a saved config. Pass -d <path>.rviz on the command line or as a launch-file argument, rather than opening RViz2 bare and re-adding displays every session:

ros2 run rviz2 rviz2 -d config/<robot>.rviz

In a Python launch file, pass the same flag through Node's arguments: arguments=['-d', PathJoinSubstitution([pkg_share, 'config', '<robot>.rviz'])]. Keep one config per robot/task (nav debugging vs. manipulation vs. sensor calibration) rather than one config trying to cover everything — a display list with everything on is slower to render and harder to read than a focused one.

The nothing-shows-up checklist. Walk these in order before assuming the publisher is broken:

1. Fixed Frame wrong or not broadcasting. Global Options → Fixed Frame must name a frame that's actually present in the TF tree right now — a red/orange warning next to a display, or the display simply staying empty, both point here first. Confirm the frame exists and is current with ros2 run tf2_ros tf2_echo <fixed_frame> <a_frame_the_data_uses> (see ros2's TF2 reference for what a stale vs. missing transform looks like) — don't touch anything else until this returns clean data.
2. QoS mismatch between the display and the publisher. Each display has a Topic section with a QoS override (Reliability: Reliable/Best Effort, Durability: Volatile/Transient Local) that must be compatible with the publisher's own QoS, not merely "set to something" — a sensor publishing Best Effort with a display left on the default Reliable will silently receive nothing, no error dialog. Check the publisher's actual QoS (ros2 topic info <topic> --verbose) and match the display's override to it; see ros2's QoS reference for the compatibility rules themselves.
3. use_sim_time mismatch or a stalled /clock. In simulation, if RViz2 (and the nodes publishing the data it's trying to show) don't agree on use_sim_time, or the sim's /clock has stalled/paused, timestamps stop advancing and time-sensitive displays (TF, anything with a message filter) stay empty even though the topic is technically publishing. Confirm ros2 topic echo /clock is advancing and every relevant node was launched with the same use_sim_time value — see nav2's sim-time directive for why this has to be set globally, not per-node.

Nav2 debugging display set. For debugging a navigation stack, add (on top of TF and RobotModel): Map (topic /map), two Costmap displays — one on /global_costmap/costmap, one on /local_costmap/costmap — a Path display for the planned path, a Polygon display for the robot footprint, LaserScan/PointCloud2 for the raw sensor(s) feeding the costmaps, and a MarkerArray for behavior-tree/ planner debug markers if the stack publishes them. Send goals with the "Nav2 Goal" tool in the toolbar. See the nav2 skill's common-failures reference for what each display should look like when the stack is healthy vs. broken (e.g. a costmap with no obstacles where there should be some).

Platform gotchas

• No headless or web mode. RViz2 needs a local X11/Wayland display — there is no server-side/remote flag. On a headless or remote box, don't attempt X11 forwarding as the default; use foxglove (see the visualization umbrella's selection table).
• macOS has no native ROS 2, so RViz2 (which links against ROS 2) runs inside Docker on a Mac dev machine along with the rest of the ROS 2 stack — see environments' and ros2's macOS gotcha for the container/display setup; this is not an RViz2-specific workaround.
• A stale config referencing a display plugin that isn't installed (e.g. a Nav2-specific display plugin missing from a minimal install) fails that one display silently or with a small error banner, not a hard crash — the rest of the config still loads. Check the Displays panel for a greyed-out/errored entry if a saved config "loses" one display after moving to a different machine.

Customization

• New robot, same task: copy the closest existing .rviz config, update the RobotModel's topic/description source and any frame-specific display settings (e.g. a LaserScan topic name) to the new robot's namespace, then re-save under a new filename rather than overwriting a config that other tasks still reference.
• Multi-robot/namespaced setups: each display's Topic field needs the fully-namespaced topic (e.g. /robot1/scan), and Fixed Frame needs the namespaced frame if TF is also namespaced — a config built for a single-robot setup will silently show nothing when pointed at a namespaced one until every topic/frame field is updated.

References

• Upstream: RViz2 documentation (direct fetch attempted on 2026-07-10; the page returned an access-control error rather than content — cross-check against the live page before relying on exact UI wording), rviz2 package page, index.ros.org (fetched directly this session — source of the distro-coverage claim above, used in place of the docs.ros.org page which did not return content), ros2/rviz GitHub repo (source of the plugin/display implementations; fetched directly on 2026-07-10 — CLI flag documentation was not present in the README itself, so the -d flag above was confirmed via search-synthesis of ROS community docs, e.g. ROS Answers on loading a specific config file and The Construct's RViz config-file walkthrough — re-verify against rviz2 --help before depending on it in a script). Sibling skills: ros2 (TF2/QoS foundations this skill applies but does not re-teach), nav2 (costmap/BT internals behind the Nav2 debugging display set), foxglove (remote/web viewing), rerun (ML/data-centric logging), visualization (umbrella, routes here).

Changelog

• 1.0.1 (2026-07-12): skill-refiner run 1 — provenance claims date-stamped ('this session' → 2026-07-10, the authoring session) so the staleness sweep can age them.