Finite-Memory CL Based on Multisensor Information Fusion Using Neural Networks for Multiple Humanoid Robots

Cooperative localization (CL) is a method for achieving accurate and reliable localization of multiple robots by sharing sensor measurements or position information through wireless communications among the robots and has been successfully applied to various mobile robots. However, in humanoid robots, swaying of the body caused by walking results in significant errors in LiDAR sensor measurements, degrading the accuracy of CL. To address this problem, we propose a novel CL algorithm aided by odometry information that is less affected by the swaying of humanoid robots. First, we design a finite-memory CL (FMCL) algorithm based on wireless sensor networks and LiDAR measurements that has a finite-memory structure, preventing the accumulation of errors in sensor measurements or models. Second, we process the odometry information with an artificial neural network (ANN) to obtain auxiliary position estimates. Third, the position estimates obtained by the two methods are integrated with another ANN to complete the odometry-aided FMCL (OAFMCL). Finally, we conduct CL experiments using multiple humanoid robots and demonstrate the superiority of the proposed OAFMCL over state-of-the-art localization algorithms.