A Rust-based ROS2 visualizer built on wgpu + egui. RViz alternative.

Each tagged release publishes Ubuntu 22.04 / ROS Humble and Ubuntu 24.04 / ROS Jazzy artifacts on the Releases page — both as a tarball and as a .deb.

The .deb declares all runtime dependencies, including the ROS message packages and Vulkan loader, so apt pulls them in for you. Pick the file that matches your distro:

curl -LO "https://github.com/OWNER/REPO/releases/latest/download/fastviz-jazzy_<version>-1_amd64.deb"
sudo apt install ./fastviz-jazzy_<version>-1_amd64.deb
fastviz --config /usr/share/fastviz/configs/default.toml

The installed binary has the ROS lib directory baked into its rpath, so you do not need to source /opt/ros/<distro>/setup.bash before launching it.

tag=$(curl -s https://api.github.com/repos/OWNER/REPO/releases/latest | grep -oE '"tag_name": *"[^"]+"' | cut -d'"' -f4)
# pick `noble-jazzy` or `jammy-humble`
asset="app-${tag}-x86_64-unknown-linux-gnu-noble-jazzy.tar.gz"
curl -L -o fastviz.tar.gz "https://github.com/OWNER/REPO/releases/download/${tag}/${asset}"
tar -xzf fastviz.tar.gz
sudo install -m 0755 "${asset%.tar.gz}" /usr/local/bin/fastviz

Replace OWNER/REPO with the repo slug (or just download the asset from the Releases page in a browser).

The release binary is dynamically linked. On the host you'll need:

• ROS2 Humble (Ubuntu 22.04) or ROS2 Jazzy (Ubuntu 24.04) — pick the artifact that matches. Other distros are untested.
• ROS2 message packages used by the subscribers (substitute humble for jazzy on 22.04):

  sudo apt-get install -y \
    ros-jazzy-tf2-msgs ros-jazzy-nav-msgs ros-jazzy-sensor-msgs \
    ros-jazzy-geometry-msgs ros-jazzy-visualization-msgs

• Vulkan loader + an ICD for wgpu:

  sudo apt-get install -y libvulkan1 vulkan-tools mesa-vulkan-drivers

  On NVIDIA hosts, the proprietary driver provides its own ICD — install nvidia-driver-* from the Ubuntu archive (or the upstream .run installer). Verify with vulkaninfo --summary.
• Windowing/input libraries (usually already present on a desktop install): libxkbcommon0, libwayland-client0, libxcb1.

sudo apt-get install -y build-essential pkg-config clang libclang-dev \
  libx11-dev libxcb1-dev libxcb-render0-dev libxcb-shape0-dev libxcb-xfixes0-dev \
  libxkbcommon-dev libwayland-dev libssl-dev \
  libvulkan1 vulkan-tools mesa-vulkan-drivers
export LIBCLANG_PATH=/usr/lib/x86_64-linux-gnu   # r2r's bindgen step needs this
source /opt/ros/jazzy/setup.bash
cargo build --release

The binary lands at target/release/app. Toolchain version is pinned in rust-toolchain.toml; rustup picks it up automatically.

If your host doesn't have ROS2 Jazzy (e.g. you're on Fedora, Arch, or a non-24.04 Ubuntu), use the bundled dev container — it ships Ubuntu 24.04 + Jazzy + the r2r build deps + the Vulkan loader, all pre-installed.

# VS Code / Cursor:
#   1. Install the "Dev Containers" extension.
#   2. Open the repo folder, then "Reopen in Container".

NVIDIA GPU passthrough requires nvidia-container-toolkit on the host. The container ships an NVIDIA Vulkan ICD manifest that activates automatically when the toolkit bind-mounts the driver libs; on non-NVIDIA hosts it falls back to Mesa.

See .devcontainer/Dockerfile for the exact image — it's also a good reference if you want to build your own container.

The .deb installs a desktop entry + icon system-wide, so a packaged install shows the fastviz icon automatically. A dev build (cargo run, including from the dev container) does not — and on Wayland the icon is resolved by the host compositor from a matching .desktop file, so installing it inside the container has no effect. Run this once on the host to get the same icon:

./icons/install-desktop.sh

It drops the same fastviz.desktop + icon into ~/.local/share, matching the window's app_id (fastviz). On an X11/XWayland session the icon already works without this, since it's sent over the wire by the app itself.

Source ROS2 first, then launch with a config file:

source /opt/ros/jazzy/setup.bash
fastviz --config configs/default.toml

Common invocations:

# TurtleBot 4 Gazebo sim — picks the URDF up off /robot_description
fastviz --config configs/turtlebot4.toml

# Override the reference frame from the CLI (wins over the config file)
fastviz --config configs/default.toml --ref-frame odom

# Load a URDF from a file directly (skips /robot_description)
fastviz --config configs/default.toml --urdf /path/to/robot.urdf

Mesh files referenced from a URDF can be .stl, .obj, or .dae (Collada). package:// URIs are resolved through AMENT_PREFIX_PATH.

• Left mouse drag — orbit
• Right mouse drag — pan
• Scroll — zoom
• F — reset view
• T — top-down view
• S — side view
• Esc — quit

fastviz [--ref-frame FRAME] [--config PATH] [--urdf PATH] [--width N] [--height N]

--ref-frame and --urdf win over the matching config-file values when both are set.

configs/default.toml mirrors RosConfig::default() and is the source of truth. Other presets live alongside it (e.g. configs/turtlebot4.toml). Key features:

• Each per-message kind ([map], [poses], [pose_arrays], [paths], [scans], [points], [markers], [marker_arrays]) takes a topics = [...] list.
• A bare "*" element enables polled discovery: anything in the ROS graph with the matching message type is auto-subscribed within ~1 s. Works for the per-message kinds above (not [map], which is single-topic).
• Per-topic QoS overrides via [<kind>.qos."<topic>"]: reliability, durability, depth.
• Visual style per kind (arrow, paths.style, scans.style, points.style).
• [tf] block lets you remap /tf / /tf_static topic names + override their QoS — handy for robots that publish under a namespace.
• [urdf] block: set either path (URDF/xacro file on disk) or topic (std_msgs/String topic carrying the URDF XML, typically /robot_description). joint_states_topic defaults to /joint_states.

Example with wildcard scans, a per-topic QoS override on /map, and a robot loaded off /robot_description:

reference_frame = "map"

[tf]
topic        = "/tf"
static_topic = "/tf_static"

[urdf]
topic              = "/robot_description"
joint_states_topic = "/joint_states"

[map]
topics = ["/map"]
[map.qos."/map"]
durability  = "transient_local"
reliability = "reliable"

[scans]
topics = ["*"]                # auto-discover every sensor_msgs/LaserScan
style  = { size = 4.0, color = [1.0, 0.95, 0.20] }

Milestone 0.5 is complete: TF, OccupancyGrid, PoseStamped, PoseArray, Path, LaserScan, PointCloud2, URDF + JointState with STL/OBJ/DAE meshes, visualization_msgs/Marker(Array), and TOML config with polled wildcard discovery + per-topic QoS. See MBABYSTEPS_1.md for the running plan.