Research on Navigation System of Agricultural Wheeled-Mobile Robot Using Multi-Information Fusion
|School||Nanjing Agricultural College|
|Course||Mechanical and Electronic Engineering|
|Keywords||Agricultural robot GPS AR model Kalman filtering Gray prediction|
Due to the complexity of agricultural environment, it is not only difficult to meet the demand of high-precision and reliable autonomous navigation of agricultural robot by using single sensor information, but also is often costly. So this paper focused on the autonoumous navigation of agricultural wheeled robot with multi-sensor information, using the algorithms of GPS error analysis, Kalman filtering and gray prediction. The main research and creative achievement included following aspects:By analying the correlation of GPS positioning error, AR models of static and dynamic errors were established using time series methods. Subsequently, the method for dealing with the positioning error in navigation was presented, and experiments were carried out with the agricultural robot. The experimental results showed that the correlation in positioning errors was decreased. The positioning error signal was similar to white noise, and the average value of positioning error was reduced to-0.0022m from 0.1951m.The multi-sensor navigation system was established with sub-meter level DGPS, AHRS M2 and photoelectric encoders, and Kalman filtering algorithm was adopted to achieve multi-information fusion. The results showed that the designed Kalman filter could smooth and stabilize the GPS positioning, and significantly reduced the effect of fluctuation of GPS datas on the navigation system. The GPS positioning accuracy was improved by 0.5-1m compared with the original data. The stable tracking error was about±0.15m, and the heading deviation was about 5°. By incorporating GPS error processing algorithms into the Kalman filtering, the GPS positioning accuracy was increased furtherly, being about 2-5m higher in straight line path tracking and 3-5m higher in headland turning, and the tracking error of headland turning was reduced from around 1.5m to 0.3m.The algorithm withχ2 checking and gray prediction successfully detected and isolated the fault caused by the GPS failure, which only depended on the valid Kalman filtered data and GPS historical data without any other sensor information. Simulation results showed its effectiveness.The proposed algorithm not only can reduce the cost of navigation system, but also ensure the adaptability and robustness. It provides a feasible mean to achieve autonomous navigation of agricultural machinery.