Comparative Analysis of SLAM Algorithms
Image 1: Yahboom ROS Master X3 Robot
Description: The Yahboom ROS Master X3 robot, equipped with an RPLIDAR sensor and Orbbec depth camera, was used for this SLAM project. For more details, visit the Yahboom ROS Master X3 website.
Image 3: Cartographer SLAM vs. Ground Truth
Description: Comparison of Cartographer SLAM's map with the ground truth, showing high accuracy in capturing the overall structure with minor discrepancies.
Image 4: GMapping SLAM vs. Ground Truth
Description: Comparison of GMapping SLAM's map with the ground truth, demonstrating good performance in simpler environments with some deviations.
This project compares the performance and mapping accuracy of two widely used Simultaneous Localization and Mapping (SLAM) algorithms, GMapping and Cartographer, for indoor robotic navigation. SLAM enables autonomous robots to simultaneously build a map of an unknown environment and estimate their position within it. GMapping uses a Rao-Blackwellized particle filter for 2D mapping, while Cartographer uses pose graph optimization and loop closure for 2D and 3D mapping.
The experimental setup used a Yahboom ROS Master X3 robot equipped with an RPLIDAR to collect mapping data across indoor environments with different levels of complexity. Mapping accuracy was quantitatively evaluated by converting generated occupancy-grid maps into physical dimensions using their pixel resolution and comparing them against manually measured ground-truth dimensions using Mean Absolute Error (MAE) and Mean Squared Error (MSE).
The key findings were:
- GMapping performed better in simpler, structured environments with fewer obstacles.
- Cartographer performed better in more complex layouts, where its graph-based optimization and loop-closure capabilities provided an advantage.
During testing, a practical limitation was identified in the sensor configuration. The RPLIDAR was mounted approximately 7 inches above the ground, creating blind spots for lower obstacles. These blind spots occasionally resulted in collisions and incomplete obstacle representation in the generated maps. A depth-camera and 2D LIDAR sensor-fusion approach was therefore proposed to improve obstacle detection across different heights.
The project demonstrates how algorithm selection, environmental complexity, sensor configuration, and evaluation methodology can influence SLAM performance in real-world indoor environments.