A Navigation Cognitive System Driven by Hierarchical Spiking Neural Network

Even equipped with expensive artificial sensors, classical simultaneous localization and mapping systems are still challenged by various problems in realistic scenarios. For achieving better performance, people strive to find alternative solutions on the brain-inspired approaches. Rodents have excellent spatial learning and memory ability. Inspired by the neural computation mechanisms of rodent spatial navigation, here we first present a novel cognitive architecture and then develop a navigation system modeling the complete information processing path from environment sensing to cognitive map building. In this hybrid cognitive system, hierarchical spiking neural networks of theta grids, grid cells and place cells encode the location information and produce the large-scale cognitive map input only by a monocular cheap camera. Experimental results show that the cognitive navigation system proposed achieves satisfactory performance on the open data from KITTI website, with the translation error 5.7%, proved that the brain-inspired approaches can work well even integrated with the artificially mono-visual cues.