Trajectory prediction and visual localization of snake robot based on BiLSTM neural network

Snake robot’s low view angle makes them vulnerable to being blocked by obstacles, which causes visual loss and track offset. In this paper, a trajectory prediction method and visual localization system for snake robots based on a Bi-directional Long Short-Term Memory (BiLSTM) neural network are proposed. First, the kinematics model of the snake robot is established by using the Denavit Hartenberg (DH) method, and the robot’s control system is described in detail. Then, the trajectory prediction model based on a BiLSTM neural network is proposed, and the network training platform is introduced. Meanwhile, the BiLSTM network’s model parameters for trajectory prediction are analyzed and optimized. To demonstrate the advantages of the proposed method, three other network prediction methods are compared. Furthermore, in order to solve the visual tracking loss of the snake robot, a novel trajectory prediction and visual localization systems based on a BiLSTM neural network and the Oriented Brief-Simultaneous Localization and Mapping (ORB-SLAM3) systems are designed. Finally, the effectiveness of the proposed method is verified by experiments.