Localizability of Laser SLAM Robot Based on Deep Learning

Abstract

Positioning is the key technology in the field of mobile robot, and it is also the basis of various tasks such as autonomous movement. Localizability is a measure of the ability to achieve accurate positioning for a given location, and it is an important indicator to further avoid positioning failure in upper tasks. Taking the fusion positioning method based on map matching and trajectory estimation as the research object, this paper analyses its working mechanism, designs different neural networks for different positioning methods, and proposes a neural network model composed of convolutional neural network, recurrent neural network and multi-layer perceptron. The model can be used for mobile robot estimate localizability. The simulation and experimental results show that the model can accurately estimate the localizability of the robot in a given map.