Research On Relative Localization Of Autonomous Land Vehicle Based On Kalman Filter

It is one of the key problems of Autonomous Land Vehicle (ALV) to locate itself by different techniques. In this thesis, the problem of Relative (local) Localization has been studied: evaluating the position and orientation using information provided by various on-board sensors. In the Autonomous Red-flag Saloon I (an ALV test bed of our research group), the on-board sensors are two encoders mounted on the rear wheels and one sensor for the rotation of the front wheel.A popular local technique, dead reckoning, employs simple "Ackman" kinematic model on odometric data to compute the position of ALV relative to its start position. Experimental results show the model is almost right while the vehicle runs on the flat road or at low speed. To introduce the information provided by the front wheel in our localization system and to show what other factors such as the pressure of the tire, the preload of the vehicle, etc, can change the parameter of the kinemic model, the dynamic model of a vehicle with four wheels has been studied. The kinematic model of the vehicle can be considered as a special case of dynamic model. From this aspect, the following results can be obtained. 1: Why the kinematic model cannot be used while the vehicle runs at high speed. 2: How to combine the information provided by the front wheel with the information provided by the rear wheel to perform the localization. 3 How the other reason such as the pressure of the tire, the preload of the vehicle, etc, affect the parameter of the kinematic model of the vehicle.The localization method of dead reckoning cannot be used for long distance and may be not robust. The kinematic model always has some inaccuracies, encoders have limited precision and there are external sources affecting the motion that are not observable by the sensors (e.g. wheel slippage). Applying kalman filter techniques can provide substantial improvement. The kalman filter provides an efficient recursive means to estimate the state of a process, in the way that minimizes the mean of the squared error. In our localization system, the Extended Kalman Filter (EKF) and the Vector Kalman Filter (VKF) are applied to the localization model of the ALV. Experiment results show a remarkable improvement in the accuracy of position and orientation provided by our localization system.