Neural Extended Kalman Filter for State Estimation of Automated Guided Vehicle in Manufacturing Environment

Abstract

To navigate autonomously in a manufacturing environment Automated Guided Vehicle (AGV) needs the ability to infer its pose. This paper presents the implementation of the Extended Kalman Filter (EKF) coupled with a feedforward neural network for the Visual Simultaneous Localization and Mapping (VSLAM). The neural extended Kalman filter (NEKF) is applied on-line to model error between real and estimated robot motion. Implementation of the NEKF is achieved by using mobile robot, an experimental environment and a simple camera. By introducing neural network into the EKF estimation procedure, the quality of performance can be improved.