Multi-source Data-fusion Robot Navigation Framework in Low-texture Environments

We propose a 2D multi-source data-fusion SLAM system with visual-aided pose estimation, structured within a flexible triple-layer navigation framework. This system is tai-lored to optimize efficiency and accuracy in robot navigation, particularly in low-texture environments. The first layer utilizes visual information from an RGB camera, employing Apriltag detection to provide visual-aided pose estimation for the 2D SLAM. The second layer comprises a configurable 2D SLAM system that integrates various data inputs, including LiDAR, IMU, and encoders. The third layer features a planning and control module that utilizes accurate environmental information to guide the robot towards its target. We validate the proposed system through extensive simulations and physical experiments. To demonstrate the effectiveness of the framework, we design a wheeled robot equipped with the complete navigation framework. Results from both simulations and experimental evaluations indicate that the proposed navigation framework achieves high accuracy and efficiency in navigating low-texture environments.