Robot Localization and Orientation Detection Based on Place Cells and Head-Direction Cells

Place cells and head-direction cells play important roles in animal navigation and have distinguishable firing properties in biology. Recently, a slowness principle has been argued as the fundamental learning mechanism behind these firing activities. Based on this principle, we extend previous work, which produced only a continuum of place and head-direction cells and mixtures thereof, to achieve a clean separation of two different cell types from just one exploration. Due to the unsupervised learning strategy, these firing activities do not contain explicit information of position or orientation of an agent. In order to read out these intangible activities for real robots, we propose that place cell activities can be utilized to build a self-organizing topological map of the environment and thus for robot localization. At the same time, the robot's current orientation can be read out from the head-direction cell activities. The final experimental results demonstrate the feasibility and effectiveness of the proposed methods, which provide a basis for robot navigation.