Self-Location Based on Grid-like Representations for Artificial Agents

Self-location plays a crucial role in framework of autonomous navigation, especially GNSS/radio-denied environment. At the current time, self-location for artificial agents still has to resort visual and laser technologies deep neural networks, which cannot model environments effectively, some dynamic complex scenes. Instead, researchers have attempted transplant navigation principle mammals into intelligence (AI) fields. As kind mammalian neuron, grid cells are believed provide context-independent spatial metric update representation self-location. By exploiting mechanism cells, we adopt oscillatory interference location encoding. Furthermore, process decoding, capacity is extended significantly wide range without phase ambiguity, based on multi-scale periodic supported by step-wise unwrapping algorithm. Compared with previous methods, proposed grid-like can achieve much wider limitation imposed scales cells. It also able suppress noise efficiently. The method validated simulation results.