Implementation of an adaptive EKF to multiple low cost navigation sensors in wheeled mobile robots

The present aim of this research is to design a navigation sensor suite for a newly built mobile robot using low cost multiple sensors. A basic requirement for an autonomous mobile robot is its ability to localize itself accurately. This paper describes an accurate method for generating navigational data for a wheeled mobile robot. An adaptive extended Kalman filter (AEKF) is used to fuse data from multiple low cost sensors. In order to estimate the spatial position of a wheeled robot, a combination of accelerometers, a rate gyroscope and two wheel encoders are used. The system discussed in this paper has more measurement sensors than system states and therefore the sensors give overlapping, low-grade information affected by noise, bias, drift, etc. The dynamics of the robot and sensor system are non-linear. Therefore an AEKF is used to estimate these overlapping low-grade measured sensor data and give the best possible estimate of the mobile robot position. The adaptive mechanism in this case uses the Riccati Equation adaption. The basic idea is to change the Kalman Gain. This is done by changing the Process noise co-variance matrix adaptively. Simulations show an improved performance in the estimates from the AEKF when compared to the EKF.