Posture Estimation for Autonomous Weeding Robots Navigation in Nursery Tree Plantations

摘要

The presented research aims at developing a sensor fusion technique for navigationalposture estimation for a skid-steered mobile robot vehicle in nursery tree plantations. RTK-GPS andFiber Optic Gyroscope sensors were used for determining the position and orientation of the robotvehicle. An Extended Kalman Filter (EKF) was developed through making the use of thecomplementary error features of these sensors. A specially designed experimental platform wasused to generate circular and linear reference trajectories for RTK-GPS calibration and errormodeling. The RTK-GPS error was modeled by an auto-regression method and error states wereincorporated into EKF design.<br><br>The EKF with AR (2) model was implemented on straight line data to check the effectiveness of thedeveloped algorithm. The mean error after incorporating AR (2) model with EKF reduced significantly with 2.63 cm and 0.37 cm in x and y direction, with standard deviation of 1.86 cm and 0.65 cm,respectively for line 1. For line 3 and 4, the mean measurement error in y direction was 9.17 cm and0.10 cm, respectively. After filtering, the error in y direction reduced more than 98%. The filter waseffective in reducing the mean errors of the system, in x and y direction for all the four lines. Further,it could also be stated that the errors were observed more in the direction of travel of the robot. Whenrobot was navigated through the poles, the positioning accuracy of the system increased afterfiltering. The accuracy of the system can further be enhanced by fine tuning of system noisecovariance matrices.