Cognitive robotics: Deep learning approaches for trajectory and motion control in complex environment

Simultaneous Localization and Mapping (SLAM) is the research hotspot of robot positioning and navigation. In a large-scale complex environment, closed-loop detection by vision or lidar has low reliability and high computational cost. To solve this problem, a graph optimization SLAM algorithm based on YOLOv5 (You Only Look Once version 5) and Wi-Fi fingerprint sequence matching is proposed. The proposed method utilizes fusion deep learning approaches to enhance the accuracy and robustness of closed-loop detection to navigate the robot. The algorithm uses an effective object detection network and the fingerprint sequence for closed-loop detection to figure out the dynamic semantic information within a scene. Therefore, the traditional matching based on fingerprint point pairs is extended to include matching of fingerprint sequences. This can greatly reduce the probability of closed-loop misjudgment, ensuring the accuracy of closed-loop detection and meeting the accuracy requirements of the SLAM algorithm in a wide range of complex environments. The proposed algorithm is verified with two sets of experimental data (the robot starts from different starting points): the accuracy of the proposed algorithm is 22.95% higher than that of the first set of data compared with the Gaussian similarity method; the second group of data increased by 39.19%. The experimental results show that the proposed method improves the accuracy and robustness of mobile robot localization and mapping.