Intro

This report looks at Kalman filters and how they can be used with the aim to improve IMU accuracy. Kalman filters are a optimal estimation algorithm used to predict the true state (a description of the system) from a set of noisy measurements. The Kalman filter uses the current state along with a model, that reflects the physics of the system, to predict the next state. This is then compared with measurements to improve on this prediction. The model in the Kalman filter is dependent on how the system behaves, whereas most filters use the same algorithm with a couple of parameters that are tuned. Often the Kalman filter will use measurements from multiple types of sensors to improve accuracy, sensor fusion.

Experiment I uses a 6 axis IMU to track position and attitude. The IMU measures acceleration, in the x, y and z directions and angular velocity in the yaw, pitch and roll directions. The position and attitude estimates are inaccurate due to drift associated with integration. Attitude estimation can be improved using a Kalman filter that fuses the accelerometer and gyroscope measurements. Position estimation can be improved by using a Kalman filter that fuses the accelerometer measurements with position measurements.

Contents

• 1 Basics Of Filtering
• 2 Kalman filters
• 3 Example: Battery output
• 4 Example: Velocity from position
• 5 Example: Attitude using a gyroscope and accelerometer
• 6 Example : Position using GPS and accelerometer data
• 7 Experiment: Attitude Using a 6/9 axis IMU
• 8 Experiment: Quantitative Comparison of Kalman Filter Performance
• Bibliography