Multi-channel Bayesian adaptive resonance associative memory for environment learning and topological map building
Wei Hong Chin, Chu Kiong Loo, Naoyuki Kubota · 2015
This paper presents a new network for environment learning and online topological map building. It comprises two layers: input and memory. The input layer collects sensory information and incrementally categorizes the obtained information into a set of topological nodes. In the memory layer, edges are connect clustered information (nodes) to form a topological map. Edges store robot's actions and bearing. The advantages of the proposed method are: 1) it represents multiple places using multidimensional Gaussian distribution and does not require prior knowledge to make it work in a natural environment; 2) it can process more than one sensory source simultaneously in continuous space during robot navigation; and 3) it is an incremental and using Bayes' decision theory for learning and inference. Finally, the proposed method was validated using several standardized benchmark datasets.