Adaptive Positioning And Navigation Method Based On Multi-source Data Fusion For Outdoor AGV

With the transformation and upgrading of China's manufacturing industry,the rapid development of smart factories and smart warehousing has an explosively growing demand on intelligent AGV,which leads to a prosperous AGV market.However,there is a phenomenon that outdoor AGVs have a large demand but a small proportion in actual applications,which is mainly restricted by some conditions such as the grand outdoor scenarios and complex environment With the improvement of sensing technology and in-depth research on positioning and navigation methods,the application field of AGV has gradually expanded from indoor environment to outdoor environment.How to ensure that AGV can accurately perceive its own status in the external surrounding environment and continue to drive safely along the planned path has become a research focus.Outdoor AGV is facing an unprecedented progressing window.The dissertation has conducted related work on the precisely positioning and stable navigation of AGV in outdoor environment.Aiming to solve the limitations and vulnerabilities of positioning and navigation system information from single-source sensor in outdoor environments and the nonlinear problem of outdoor pose estimation for outdoor AGV,the positioning method with multi-sensor information fusion under complex outdoor environment is studied.The cubature Kalman filter algorithm is used as the basic fusion algorithm,and the estimated value of the AGV pose by dead reckoning method and the measured value from GPS and IMU are calculated to obtain the optimal pose estimation through the weighted Kalman gain coefficient.In the process of positioning iteration,the CKF error covariance square root matrix simplified by QR decomposition replaces the ill-conditioned matrix containing the accumulated positioning errors caused by word length limitation,by which the problem of location instability or even divergence arisen from the ill-conditioned covariance matrix for outdoor navigation is solved.based on time series,a fusion positioning algorithm using time sequential filtering method for sensing data sampled with different periods is proposed,making full use of sensor observation information.The QR square root filtering method is used to ensure the positioning stability,the time sequential filtering method to improve the availability of sensor information,and therefore the accurate location of outdoor AGV with multi-sensor data fusion under stable status is realized.Aiming at the problem that the uncertain factors of outdoor environment would affect the robustness of outdoor AGV positioning,an adaptive strong tracking positioning method under outdoor uncertain conditions is studied.An improved Sage-Husa adaptive filtering algorithm is adopted,in which the measurement noise is estimated and adjusted in real time by the residual updating strategy as well as monitored simultaneously.A biased estimation method is used to recalculate covariance matrix when it is non-positive definite,which can ensure the stability of the positioning filter and adaptively adjust noise parameters in real time to improve positioning accuracy while the system model is uncertain.A strong positioning tracking algorithm based on multiple fading factors which are calculated according to influences of various types of sudden changes during driving process on pose parameters corrects the error covariance matrix deduces the Kalman gain coefficient,and then the weighted observed values are obtained,by which the divergence problem of long-term positioning data caused by unpredictable factors after a sudden change in the pose of outdoor AGVs can be handled.The adaptive strong tracking algorithm improves the positioning robustness under the condition of uncertain model parameters,enhances the reliability and timeliness of positioning and tracking,and can stably and continuously provide absolute positioning information for an AGV.Considering that it is difficult to get navigation deviation during curve movement due to the lack of relationship between the absolute positioning information of the AGV and the driving path,and how to accurately track the path without the absolute pose information,the navigation deviation is calculated by transforming the absolute pose in the Cartesian coordinate system to Frenet coordinates.When the GPS is losing lock in the outdoor environment,the redundant navigation information from an on-board camera is made use of to perceive the driving road conditions.After image correction to barrel distortion the image is transformed into HSV color space.Due to high saturation and strong contrast the path objective is separated and the navigation information is emphasized.The image preprocessing of morphological filtering and bilateral filtering removes road texture and background noise.Edge detection and Hough line detection are used to extract the corner coordinates of the path area.The inverse perspective transformation matrix is calculated according to the corner points of the path area of the reference image,and the relationship between the image coordinates and the actual position is established to derive the navigation deviation,so as to realize the real-time and accurate acquisition of the navigation deviation when the GPS is losing lock.A case study simulation of fuzzy PID deviation correction control of path tracking verifies that the proposed method can effectively ensure the path tracking correctness and driving safety of outdoor AGVOutdoor AGV positioning and navigation experiments have been carried out.Multi-source heterogeneous information is acquired by the on-board perception system(RTK-GPS,IMU,camera,vision sensor),and an improved strong tracking adaptive cubature Kalman filtering algorithm based on QR decomposition is programmed with C++code under the WIN system to perform the AGV optimal pose estimation.By calibrating the GPS static positioning accuracy and setting a two-dimensional code positioning tag according to a given distance,an outdoor absolute positioning datum with a position accuracy of ± 1mm and an azimuth accuracy of ± 1° is established.The effectiveness and feasibility of the multi-sensor information fusion adaptive positioning method proposed is verified through outdoor dynamic linear and curved movement experiments without prior sensor noise data.The position location error is less than 5cm,and the standard deviation is less than 4cm.Match the measured value of the reference position path recognition area with the reference image.The inverse perspective transformation matrix is calculated,and the image coordinates are projected into the actual road model to calculate the navigation deviation.The positioning accuracy of visual navigation is less than 2cm,and the orientation positioning accuracy is less than 0.5°,which meets the accuracy requirements of path navigation deviation information in the case of GPS losing lock.In summary,the outdoor adaptive positioning integrated navigation algorithm proposed in the dissertation can implement dynamic centimeter-level positioning accuracy.When GPS loses lock,centimeter-level navigation deviation data is supplemented based on road image information,which provides reliable positioning and navigation information for outdoor AGVs.