Modified Cascade Kalman Filter for Sensor Data Fusion in Micro Aerial Vehicle∗

Precise attitude estimation is important for navigation, guidance and control of Micro Aerial Vehicles (MAV) as they are mostly equipped with low integrity sensors due to the constraints on MAV payload and available power. The MAV sensors such as accelerometer and magnetometer are prone to noise as they are placed in proximity to the motor due to constraints on centre of gravity (CG). Data from a single sensor are not reliable for all operating points of flight envelope as the motor vibration and magnetic flux due to the motor vary with RPM. Hence accurate attitude estimation of MAV is achieved through multisensor data fusion. In this paper, a modification to the classical cascade Kalman filter is proposed. Modified Cascade Kalman Filter (MCKF) has better estimation performance as it is a single Kalman filter similar to measurement fusion technique and also it restores the flexibility and computational efficiency of state vector fusion method. A numerical example is presented wherein the derived MCKF is implemented for the attitude estimation of MAV in the 6 DOF simulation model developed in MATLAB/SIMULINK and also for a set of calibrated accelerometer and magnetometer sensor data with motor noise acquired from the autopilot mounted on the rate table of the motion simulator. It was found that MCKF exhibits substantially improved performance compared to extended Kalman filter with only accelerometer data with motor noise.