Mammalian Brain Inspired Localization Algorithms with von Mises Distributions
share
Biological agents still outperform the artificial counterparts in navigating the first-visited environments, even with the advance of deep neural networks nowadays. To bridge this gap, by taking the localization problem as the initial step, we investigate the localization principles in mammalian brains to establish the common localization framework in both biological and artificial systems. Furthermore, inspired by the grid cells discovered in mammalian brains, a localization algorithm with circular representation is proposed. Compatible with bearing-and-distance measurement, the proposed algorithms avoid the linearization inconsistency which remains a severe problem in conventional extended Kalman filter algorithms. As the effectiveness of the proposed algorithms shown in simulation results, this paper indicates a novel localization method that is promising to further tackle the simultaneous localization and mapping (SLAM) problem.
READ FULL TEXT