Research On Multi-sensor Fusion Localization Technology Based On Adaptive Fuzzy Inference

In hydraulic engineering,pressure steel pipes are typically maintained using manually built scaffolding,which consumes significant human and material resources and poses safety risks.The process can be automated by using a mobile climbing robot equipped with maintenance tools,which requires implementing the robot’s self-localization function.This study focuses on the characteristics of working conditions in pressure steel pipes and the sensor performance degradation caused by the robot’s motion,and investigates the multi-sensor fusion localization based on wheel encoders,inertial measurement unit,and monocular camera.The hardware is integrated into the robot operating system through the driver layer,and the intrinsic calibration of the sensors as well as the extrinsic calibration between the multiple sensors are completed.Based on the analysis of the sensor’s own characteristics and actual working conditions,the accelerometer in the inertial measurement unit is abandoned,and the gyroscope is combined with the wheel encoder to form a wheeled inertial odometer.Two measurement models are constructed for the gyroscope and wheel encoder respectively,and 6 degree-of-freedom robot pose calculation is achieved.For the monocular camera,a visual odometer based on optical flow is constructed.In the feature point matching process,random sample consensus algorithm is used for outlier rejection,and the scale of the monocular camera is fixed according to the hardware platform characteristics to avoid scale drift.The inter-frame pose transformation of the robot is calculated iteratively using the closest point.The two odometers are fused using the error state Kalman filter framework.The prediction equation of the filter is provided by the wheeled inertial odometer,and the observation equation is corrected by the visual odometer.The noise in the filter is assumed to be Gaussian white noise,but in actual working conditions,situations such as tire slip do not conform to this assumption.To solve this problem,an adaptive fuzzy inference system is constructed.The left-to-right wheel speed ratio of the wheeled inertial odometer,the Z-axis angular velocity of the gyroscope,the reprojection error and the number of feature points of the visual odometer are selected as input variables for fuzzy inference to adjust the covariance in the filter.The hyperparameters in the fuzzy inference system are optimized using a learning set.The localization system is validated by conducting precision tests on an actual vehicle in an experimental field,verifying the effectiveness of the sensor fusion model.