Enhanced Visual SLAM and Path Planning for Autonomous Navigation of Wheeled Mobile Robots.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Traditional visual odometry faces localization challenges due to feature point distribution.
• Conventional SLAM systems often produce sparse point cloud maps, limiting environmental representation.
• Path planning in mobile robots requires efficient and smooth trajectory generation.

Purpose of the Study:

• To enhance localization accuracy and stability in visual SLAM.
• To improve the completeness and detail of 3D environmental mapping.
• To optimize path planning and enable real-time obstacle avoidance for mobile robots.

Main Methods:

• An enhanced visual odometry approach combining Efficient Perspective-n-Point (EPNP), Iterative Closest Point (ICP), and quadtree-based feature management.
• Dense point cloud reconstruction using RGB-D data.
• An enhanced Rapidly-exploring Random Tree (RRT) algorithm with adaptive step-size, goal biasing, and B-spline smoothing, integrated with the Timed Elastic Band (TEB) algorithm for obstacle avoidance.

Main Results:

• Significantly improved localization accuracy and stability compared to traditional methods.
• Generation of dense and detailed point cloud maps, overcoming sparsity issues.
• Enhanced path quality, computational efficiency, and real-time obstacle avoidance capabilities.

Conclusions:

• The proposed integrated system offers robust and efficient autonomous navigation for wheeled mobile robots.
• The advancements in visual SLAM, mapping, and path planning contribute to practical robotic applications.
• Real-world tests confirm the system's efficiency, robustness, and applicability on a Robot Operating System (ROS) platform.