A comparison of dynamic and kinematic observers

Abstract

The automotive industry is continuously improving the safety and comfort of the vehicles. To improve the safety and comfort of the vehicle more focus is put on improving driver assistance systems and also on making the vehicles more autonomous. In order for the safety systems to work properly and also to drive autonomously, it becomes essential that the vehicle has accurate knowledge of all the states of the vehicle. Two of such states are the roll and pitch angle. They are used to correct the images of a stereo camera, in roll control systems and for comfort assessment in autonomous vehicles.

Measuring the roll and pitch angle is in practice problematic. Gyroscopes are mostly used to calculate the roll and pitch angle by integrating the angular rates with respect to time. If low quality gyroscopes are used, this leads to inaccurate calculation of the roll and pitch angle, because the measurements contain noise or bias accumulated during the integration process. To use these low quality gyroscopes, they need to be incorporated in an robust algorithm together with other sensors. Only then the roll and pitch angle can be estimated accurately.

In this thesis a dynamic observer is developed, which has an internal vehicle dynamic model that calculates the tyre forces using an exponential tyre model. The performance of the observer is not only assessed using the true roll and pitch angle, but also compared against the performance of a kinematic observer.

To test the performance of both observers various test manoeuvres were executed in IPG CarMaker, which is a highly advanced vehicle simulation environment. The manoeuvres consisted of rapid steering inputs and hard braking to execute large roll and pitch angles. Sensor noise was added to the inputs of the observers to simulate low quality sensors. In this thesis it has been shown that incorporating dynamics into an observer significantly increases its estimation performance and outperforms the kinematic observer during all manoeuvres when sensor noise was added to the inputs of the observers.

The contribution of this thesis to the scientific field is that an exponential tyre model is used for the calculation of the roll and pitch angle in the internal vehicle dynamic model of the dynamic observer. Another contribution is that this thesis not only compares the performance of the dynamic observer against the real roll and pitch angles, but also against a kinematic observer, to see what the estimation improvement is when including dynamic relations.