RTAB-SLAM with IMU Integration for Indoor Mapping

A ROS2 implementation of RGB-D SLAM using RTAB-MAP with Intel RealSense D435 camera and VectorNav VN-100 IMU for indoor environment mapping and localization.

Project Overview

This project implements Real-Time Appearance-Based Mapping (RTAB-MAP) for 3D reconstruction and simultaneous localization and mapping (SLAM) in residential indoor environments. The system combines RGB-D visual data with IMU measurements to achieve robust pose estimation and dense point cloud reconstruction.

Authors: Kiran Sairam Bethi Balagangadaran, Abhinav Vaddiraju, Prasath Saravanan
Institution: Northeastern University
Course: EECE5554 - Robot Sensing and Navigation
Date: December 2024

Key Features

• Dense 3D Reconstruction: Point cloud mapping with 150-250 tracked features in textured regions
• Loop Closure Detection: Bag-of-words based approach with 4/6 successful detections
• Low Drift: 2-3% translational drift over traveled distance
• Real-Time Performance: 40-60ms processing time per frame at 15Hz
• Sensor Fusion: RGB-D camera with IMU integration for improved pose estimation

Hardware Requirements

Required Components

• Intel RealSense D435 RGB-D Camera (1280×720, 30Hz, 0.3-3m range)
• VectorNav VN-100 IMU (9-DOF, 100Hz native)
• Computing Platform:
  • CPU: Intel i7-13750H or equivalent
  • GPU: RTX 4070 or equivalent (optional, CPU-bound operation)
  • RAM: 16GB minimum
  • USB 3.2 ports (dedicated controller recommended)

Custom Mounting

• 3D-printed bracket for rigid camera-IMU mounting (design files included)
• Ensures fixed spatial relationship between sensors for accurate calibration

Software Requirements

Operating System

• Ubuntu 24.04 LTS (recommended)
• Ubuntu 22.04 LTS (compatible)

Dependencies

# ROS2 Installation (Jazzy for Ubuntu 24.04)
sudo apt install ros-jazzy-desktop

# RTAB-MAP
sudo apt install ros-jazzy-rtabmap-ros

# RealSense SDK
sudo apt install ros-jazzy-realsense2-camera

# VectorNav ROS2 Driver
# Clone from: https://github.com/dawonn/vectornav

# Additional Tools
sudo apt install ros-jazzy-tf2-tools
sudo apt install ros-jazzy-rqt-*

Installation

1. Clone the repository:

cd ~/ros2_ws/src
git clone https://github.com/Kiran1510/RTAB-SLAM-Implementation.git
cd ~/ros2_ws
colcon build
source install/setup.bash

2. Verify sensor connections:

# Check RealSense
rs-enumerate-devices

# Check VectorNav (replace /dev/ttyUSB0 with your device)
ls -l /dev/ttyUSB*

3. Configure udev rules for VectorNav:

sudo usermod -a -G dialout $USER
# Log out and back in for group changes to take effect

Usage

Quick Start

Launch RTAB-MAP with sensors:

ros2 launch rtabmap_ros rtabmap.launch.py \
  rtabmap_args:="--delete_db_on_start" \
  depth_topic:=/camera/depth/image_rect_raw \
  rgb_topic:=/camera/color/image_raw \
  camera_info_topic:=/camera/color/camera_info \
  approx_sync:=true \
  frame_id:=base_link \
  queue_size:=30

Recording Data

Capture rosbag for offline processing:

ros2 bag record -o indoor_mapping \
  /camera/color/image_raw \
  /camera/depth/image_rect_raw \
  /camera/color/camera_info \
  /imu/data

Visualization

Launch RViz with RTAB-MAP visualization:

rviz2 -d $(ros2 pkg prefix rtabmap_ros)/share/rtabmap_ros/launch/config/rgbd.rviz

View real-time point cloud:

• Add PointCloud2 display
• Set topic to /rtabmap/cloud_map
• Set Fixed Frame to map

Configuration

Camera Settings

Located in config/realsense_config.yaml:

camera:
  resolution: 1280x720
  fps: 15
  depth_range: [0.3, 4.0]
  enable_ir_emitter: true
  enable_auto_exposure: true

IMU Settings

Located in config/vectornav_config.yaml:

imu:
  output_rate: 15  # Hz (downsampled from 100Hz)
  use_hardware_timestamps: true

RTAB-MAP Parameters

Key parameters in launch file:

• approx_sync: true - Enable approximate time synchronization (60ms tolerance)
• queue_size: 30 - Message buffer size
• Rtabmap/DetectionRate: 1.0 - Loop closure detection frequency
• Vis/MinInliers: 15 - Minimum feature matches for odometry
• GFTT/MinDistance: 7 - Feature detector spacing

Project Structure

RTAB-SLAM-Implementation/
├── config/
│   ├── realsense_config.yaml
│   ├── vectornav_config.yaml
│   └── rtabmap_params.yaml
├── launch/
│   ├── rtabmap_rgbd.launch.py
│   └── sensors.launch.py
├── src/
│   ├── timestamp_diagnostic.cpp
│   └── sensor_sync_node.cpp
├── docs/
│   ├── RTAB_Report_RSN.pdf
│   └── hardware_setup.md
├── cad/
│   └── camera_imu_mount.stl
├── results/
│   ├── point_cloud_map.png
│   └── trajectory_visualization.png
└── README.md

Results

Achieved Performance

• Mapping Quality: Dense 3D reconstruction with clear room geometry
• Translational Drift: 2-3% of traveled distance without loop closures
• Rotational Drift: 1-2° per 360° rotation
• Loop Closure Success: 4/6 intentional loops detected (0 false positives)
• Feature Tracking: 150-250 features in textured areas, 50-80 in plain regions
• Processing Speed: Real-time at 15Hz (40-60ms per frame)

Known Limitations

• Depth Quality: Degrades beyond 2.5m and in bright lighting
• Frame Dropping: Variable publishing rate (2.7-5.7Hz actual vs. 15Hz target)
• Featureless Regions: Tracking loss in plain walls/uniform surfaces
• USB Bandwidth: Frame corruption under heavy data load

Troubleshooting

Frame Dropping Issues

Problem: RTAB-MAP warnings about missing frames

[WARN] [rtabmap]: Frame dropped (X frames since last update)

Solutions:

1. Reduce camera resolution: 1280×720 → 640×480
2. Lower frame rate: 30Hz → 15Hz
3. Use dedicated USB 3.2 controller
4. Increase approx_sync tolerance to 60-80ms

Timestamp Synchronization

Problem: IMU and camera timestamps misaligned

Solutions:

1. Enable hardware timestamps for both sensors
2. Check clock source synchronization
3. Increase queue_size parameter
4. Monitor with: ros2 topic hz /camera/color/image_raw

Poor Tracking in Featureless Areas

Problem: Odometry fails on plain walls

Solutions:

1. Add visual markers or textured objects to environment
2. Reduce movement speed (< 0.3 m/s)
3. Increase GFTT/QualityLevel parameter
4. Enable IMU integration for better pose prediction

Demo & Resources

• Video Demonstration: Real-time Mapping Process
• Technical Report: docs/RTAB_Report_RSN.pdf
• Hardware Setup Guide: docs/hardware_setup.md

Future Improvements

I have acquired a Zed 2i Stereo Depth Camera with an inbuilt IMU, which eliminates the need for the RealSense and VN-100 jig/contraption. This will also make coordinate frame transforms much more simplified and straightforward. I will be doing this same implementation on the Zed Camera, with hopefully better results this time.

• Hardware GPIO triggering for perfect sensor synchronization
• Implement tighter loop closure detection (reduce missed loops)
• Add depth image preprocessing to handle bright lighting
• Optimize memory management for longer mapping sessions (>10 minutes)
• Integrate wheel odometry for better drift estimation
• Deploy on mobile robot platform for autonomous navigation

Citation

If you use this work in your research, please cite:

@techreport{rtabslam2024,
  author = {Bethi Balagangadaran, Kiran Sairam and Vaddiraju, Abhinav and Saravanan, Prasath},
  title = {RTAB-SLAM with Integration of IMU Data for Indoor Mapping},
  institution = {Northeastern University},
  year = {2024},
  type = {Course Project Report},
  number = {EECE5554}
}

References

1. M. Labbé and F. Michaud, "RTAB-Map as an Open-Source Lidar and Visual SLAM Library," Journal of Field Robotics, 2019
2. Intel RealSense D435 Product Datasheet
3. VectorNav VN-100 IMU Documentation

License

This project is released under the MIT License. See LICENSE file for details.

Acknowledgments

• Dr. Kris Dorsey and EECE5554 teaching staff at Northeastern University
• RTAB-MAP development team
• ROS2 and Intel RealSense communities

Contact

Kiran Sairam Bethi Balagangadaran
Email: bethi.k@northeastern.edu
GitHub: @Kiran1510

---

Last Updated: December 2025