Bionic SLAM Algorithm Based on Multi-Scale Grid Cell to Place Cell

Abstract

: Aiming at the problems of low positioning accuracy and angle drift in the process of simultaneous localization and mapping (SLAM), inspired by the spatial cognitive mechanism of mammalian hippocampus, a bionic SLAM algorithm for constructing information conversion from multi-scale grid cell to place cell is proposed. Firstly, the proposed algorithm introduces head direction cell and stripe cell to perceive their own motion information while generating a multi-scale grid cell to cover the entire spatial environment, which can reduce the cu-mulative error due to angular offset. Secondly, as for the problem of low localization accuracy, the proposed algorithm uses a competitive neural network under Hebb learning rule to establish the information conversion relationship from multi-scale grid cell to place cell. Meanwhile, the mapping relationship between place cell and dif-ferent landmarks in the spatial environment is constructed. Finally, the place cells with the maximum discharge rate are selected in order to form spatial cognitive topological map while realizing the autonomous localization of mobile robots. Compared with RatSLAM and ORB-SLAM2 on the KITTI public dataset, the results show that the proposed algorithm can realize autonomous localization and mapping in unknown environments by encoding the location information, while controlling the translation error at no more than 1.50 m and the rotation error at no higher than 1.0°.