Visual SLAM Guide — Intel RealSense D435i + RTAB-Map

Hardware & Software Requirements

Component	Details
Camera	Intel RealSense D435i (USB 3)
ROS 2	Jazzy/Humble
SLAM	RTAB-Map 4.56
Packages	realsense2_camera, rtabmap_ros, imu_filter_madgwick

---

System Architecture

flowchart LR
    subgraph cam["RealSense D435i"]
        rgb["RGB stream 640×360 @ 30 Hz"]
        depth["Depth stream aligned to RGB"]
        imu["IMU accel + gyro"]
    end

    rgb    --> odom["rgbd_odometry"]
    depth  --> odom
    imu    --> madgwick["imu_filter_madgwick"]
    madgwick --> imu_data["/imu/data"]

    odom     --> rtabmap["rtabmap"]
    imu_data --> rtabmap

    rtabmap --> db["~/.ros/rtabmap.db"]
    rtabmap --> rviz["RViz2"]

Loading

---

One-Time Setup

1. Install dependencies

sudo apt install ros-jazzy-realsense2-camera \
                 ros-jazzy-rtabmap-ros \
                 ros-jazzy-imu-filter-madgwick \
                 cloudcompare

2. Verify camera is detected

rs-enumerate-devices
# Must show: Intel RealSense Depth Camera 435i

If not detected: unplug and replug the USB 3 cable (blue port), then retry.

3. Build the package

cd ~/path/to/Visual-Slam/ros2_ws

colcon build --packages-select point_cloud anubi_slam
source install/setup.bash

---

Running SLAM

Fresh map (deletes previous)

source ~/path/to/Visual-Slam/ros2_ws/install/setup.bash
ros2 launch anubi_slam slam.launch.py

Resume mapping (keeps existing map)

ros2 launch anubi_slam slam.launch.py delete_db:=false

What starts automatically

Node	Role
realsense2_camera_node	Streams RGB, aligned depth, IMU
imu_filter_madgwick_node	Fuses accel + gyro → quaternion
rgbd_odometry	Estimates camera motion frame-by-frame
rtabmap	Builds map, detects loop closures, saves to DB
rviz2	Live visualization

---

Mapping Best Practices

Check odometry quality first

Watch the terminal for:

Odom: quality=87 ...   ← good, start moving
Odom: quality=0  ...   ← tracking lost, stop and recover

Wait for quality > 20 before moving.

Camera tips

Do	Don't
Point at textured surfaces (books, objects, walls with posters)	Point at blank walls or ceilings
Move slowly (~0.3 m/s)	Make fast rotations
Keep 70%+ overlap between consecutive views	Jump or shake the camera
Return to start position to trigger loop closure	Cover the lens

Loop closure

When you return to a previously seen area, a red edge appears in the RViz MapGraph panel. This corrects accumulated drift and is essential for an accurate map.

---

Saving the Map

The map is saved automatically to ~/.ros/rtabmap.db when you press Ctrl+C.

To also export as a point cloud file:

# Export .ply (default output: ~/.ros/rtabmap_cloud.ply)
rtabmap-export --cloud ~/.ros/rtabmap.db

# Export with colour
rtabmap-export --cloud --texture ~/.ros/rtabmap.db

# Save to a custom path
rtabmap-export --cloud --output /path/to/my_map.ply ~/.ros/rtabmap.db

To save the map to a specific database path:

# Set custom DB path in launch
ros2 launch anubi_slam slam.launch.py
# Then edit slam.launch.py and add to rtabmap parameters:
# 'database_path': '/path/to/my_map.db'

---

Viewing the Map

Option A — RTAB-Map database viewer (recommended, already installed)

rtabmap-databaseViewer ~/.ros/rtabmap.db

Inside the viewer:

• Edit → View 3D Map — full coloured point cloud
• View → Graph — pose graph with loop closure edges
• Edit → Export Clouds — export to .ply/.pcd

Option B — CloudCompare (best for point cloud inspection)

CloudCompare ~/.ros/rtabmap_cloud.ply
# or the lightweight viewer:
ccViewer ~/.ros/rtabmap_cloud.ply

Useful CloudCompare tools:

• Edit → Colors → Height Ramp — colour by height
• Tools → Statistics — point density, noise analysis
• Edit → Subsample — reduce point count for performance

Option C — RViz2 live view (during mapping)

RViz2 opens automatically with the launch file. Key displays:

Panel	Shows
MapCloud	Growing 3D coloured map
MapGraph	Pose graph — blue=sequential, red=loop closure
LocalCloud	What the camera sees right now
Trajectory	Camera path (green line)
OccupancyMap	2D top-down floorplan
TF	Coordinate frames — smooth motion = good tracking
RGBImage	Live colour feed

---

TF Tree (for debugging)

ros2 run rqt_tf_tree rqt_tf_tree

Expected structure:

map
└── odom
    └── camera_link
        ├── camera_color_optical_frame
        ├── camera_depth_optical_frame
        └── camera_imu_optical_frame

If map → odom is missing, RTAB-Map has not initialised yet (wait for quality > 0).

---

Troubleshooting

quality=0 — odometry tracking fails

• Point camera at a textured scene (not a blank wall)
• Move more slowly
• Lower Vis/MinInliers in slam.launch.py (currently set to 10)

No RealSense devices were found

# Reload udev rules
sudo udevadm control --reload-rules && sudo udevadm trigger
# Unplug and replug the camera, then retry
rs-enumerate-devices

source install/setup.bash warns about missing realsense path

touch ~/path/to/Visual-Slam/ros2_ws/src/realsense-ros/COLCON_IGNORE
rm -rf ~/path/to/Visual-Slam/ros2_ws/install/realsense2_camera
colcon build --packages-select anubi_slam

Build fails with No module named 'catkin_pkg'

~/path/to/Visual-Slam/ros2_ws/src/myenv/bin/pip install catkin_pkg lark empy
rm -rf ~/path/to/Visual-Slam/ros2_ws/build/point_cloud
colcon build --packages-select point_cloud anubi_slam