Abstract
This paper reports on a simulation feasibility study to investigate the direct measurement of ground plane vehicle response state variables (velocities over the ground) using GPS (Global Positioning Satellite) compensated INS (inertial instrumentation). Specifically, simulations will be run for a vehicle negotiating a test case manoeuvre where an extended KF (Kalman filter) will be designed using either state variables assumed to be derived using estimated states from a limited set of directly measured state variables (as is the current practise) or measured directly using GPS compensated INS. The estimated outputs (set up to be the relevant state variables of the vehicle response) from the extended KF will then be used by a rear wheel steer controller to effect the response of the vehicle whilst negotiating the test case manoeuvre.
There will be three separate vehicle models in each simulation test case. These being the plant model representing the vehicle itself and the predictor and corrector models within the KF design algorithm. The basis for each model will be a five degree of freedom lumped parameter (bicycle) transient vehicle model. It will be possible in the simulation to introduce plant uncertainty by using the non-linear Fiala tyre model in the plant model whilst using a linear approximation for the tyre properties in the KF design algorithm. Additionally, the ability of the extended KF to deal with non-stationary measurement noise and disturbance will be tested by the inclusion of a program switch to compare the KF performance where the noise and disturbance has both stationary and time varying mean and variance values (noise and disturbance signals to be assumed Gaussian white noise in nature). The simulation set will consist of the same transient cornering manoeuvre where the vehicle will accelerate up to a constant speed and then start a steady left hand turn. This manoeuvre was chosen to exercise the extended KF by implementing its design for a range of linearisation operating points and because the form of the response for the vehicle is easily predictable. The driver model for this work consisted of PID (proportional, integral and derivative) longitudinal and lateral feedback controllers acting on the error between the forward velocity and the yaw position in vehicle body centred coordinates and their respective demand values.
The intention behind this work was to set up a simulation architecture to enable a range of vehicle models and controller algorithms to be plugged into the transient time domain vehicle simulation environment to demonstrate and develop time varying noise and disturbance input extended KF design algorithms. The results in this paper demonstrate that such a simulation environment has been created and that the ground work has been laid for its use with full order twin track vehicle models both in the plant model and KF estimator design algorithm and with other vehicle dynamic system controllers. The eventual aim of this work is to take the Kalman estimators designed in this simulation environment and implement them in real time on an actual test vehicle instrumented with a GPS compensated INS system.
Keywords: Extended Kalman Filter Dynamic Systems