Development and evaluation of a real-time error state Kalman Filter for localization of an indoor robot

Abstract

An indoor scenario for robots has the advantage of running in a controllable environment, while the blocking of Global Navigation Satellite System (GNSS) signals makes it difficult to provide accurate absolute measurements in real-time. A robotic total station (RTS) can track a prism in real-time with millimeter-level positional accuracy. This thesis implements a real-time error state Kalman filter (ESKF), which uses an RTS, Inertial Measurement Unit (IMU), and odometry on a robot. In this thesis, the following problems are addressed: initialization of the filter, design of measurement functions for each sensor, and the real-time challenges. Two versions of this filter are implemented: one in MATLAB for theoretical proof and simulation, and one in ROS for real-time performance. Finally, a real-time result is presented and evaluated in three aspects: robustness, runtime, and accuracy.