Bionic Multiscale Grid Cell Model for Robot Localization and Navigation

Abstract

The bionic localization method based on animal spatial cognition and positioning mechanism provides a new research direction for mobile robot navigation. It also provides the possibility to break through the limitations of existing mobile robot navigation. This paper studied the construction of grid cell model for robot spatial localization, and built the training environment of rat brain spatial cell network model. The grid cell model is achieved, and the network model is trained by Adam optimization algorithm, which learned the spatial cell activities under different parameters. According to the results of space cell simulation, a self-learning multi-scale space cell model is established. In the experiment, firstly, the space cell model training set is collected and synchronized to ensure that the speed and pose information of the time synchronized robot in the experimental environment can be collected. After completing the training of the spatial cell model, we obtained the spatial cognitive activity map of the robot for the experimental environment.