Sensor Data Fusion for Positioning of Agriculture Mobile Robot using Dempster-Shafer Method

In recent year, using smart systems in agriculture in order to save costs, increase the production per unit area, minimize the hard working conditions as well as dangerous and long works, and also the precise control and supervision is unavoidable in the modern agriculture. The positioning of a mobile agriculture robot with any kind of structure and working role is one of the most fundamental and essential issues in the area of agricultural machines, and it is also a prerequisite of movement for any kind of mobile system in the farm. Therefore, this positioning always faces challenges and also is gotten a lot of attention from scholars working in this field of study. Sensor data fusion from several information sources and using various data fusion methods gives us a general precise image of the agriculture robot’s position. The Dempster-Shafer theory is one of these mentioned methods which benefits from a better performance compared with other data fusion methods, regarding the variable and unspecified workspace of agricultural robots. In this study, the methods of Dempster-Shafer and Kalman filter were used as two major tools of positioning sensors fusion related to an agriculture controllable tractor, in order to achieve the best estimation of the positioning, regarding the environmental conditions. So as to use Dempster-Shafer method in the fusion of numerical data of global positioning system (GPS), inertial measurement unit (IMU) and wheel (shaft) encoder sensors, the data reliability of each sensor is firstly determined by the standard deviation of data for each last n generated data. Then, the weighting is accomplished by the Shannon entropy method. In the simulation section, the dominant geometric equations of the studied tractor are extracted, and a proportional integral derivative (PID) controller is used in order to kinematic control of the robot. Afterward, the simulation process is run in Sim-mechanics MATLAB software. Finally, the performance of two investigated methods in this work is assessed and then compared by addition of different noises into the data of each sensor.