The mission of the future parcel delivery will be performed by UAV. However, GNSS localization in urban area experiences the notorious multipath effect and non-line-of-sight (NLOS) reception, which could potential generate about 50 meters of positioning error. A solution to guide UAV operation is to plan an optimal route that smartly avoiding the area with strong multipath effect. To achieve this goal, the impact of multipath effect in terms of positioning error at different locations must be predicted. This research proposes to simulate the reflection route by ray-tracing technique with the aids of predicted satellite positions and the widely available 3D building model. By reconstructing the multipath-biased pseudorange, the predicted positioning error could be obtained using least square positioning method. Finally, the predicted GNSS error distribution of a target area can be further constructed. New A* path planning algorithm is developed to combine with the GNSS error distribution.
Proposed flight procedure to deliver a parcel by an autonomous quadcopter.
Demonstration of the prediction of a 2D GPS positioning error map
Details of Generating GNSS Localization Error Map can be found in the video below
Conventional and proposed A* path planning algorithm based on positioning error map. Obstacles (buildings) are constructed as white area. The color bar denotes the positioning error in meters.
Due to the increase of civil applications using quadcopters, a commercial flight control system such as Pixhawk becomes a popular solution to provide the sensing and control functions of the UAV. Low-cost GNSS receiver plays an essential role in the low-cost flight control system. However, the accuracy of GNSS positioning is seriously degraded by the notorious multipath effect in mega-urbanized cities. The multipath effect cannot be eliminated but to mitigate, hence, GNSS/inertial navigation system (INS) integrated navigation is a popular approach to reduce this error. This study proposes an adaptive Kalman filter adjusting the noise covariance of GNSS measurements under different circumstances. The adaptive tuning is based on a proposed accuracy classification model trained a supervised machine learning method. Firstly, the principle component analysis is employed to identify the significant GNSS accuracy related features. Then, the supervised machine learning model is trained based on a random forest learning algorithm with the labelled real GNSS dataset covering most of scenarios in modern urban areas. To reduce the cases of miss-classifying GNSS accuracy, a fuzzy logic algorithm is designated to consider the GNSS accuracy propagation. Besides, the process noise covariance of INS is determined using Allan variance analysis. The positioning performance of the proposed adaptive Kalman filter is compared with both a conventional Kalman filter and the positioning solution provided by the commercial flight control system, Pixhawk 2.
(a) Self-assembled quadcopter; (b) Pixhawk 2 flight control system (with an autopilot software); (c) u-blox NEO-M8N GNSS chip; (d) MEMS IMU including LSM303D integrated accelerometers/magnetometers and L3GD20 gyroscopes.
Demonstration of GNSS positioning error in urban areas (a) without and (b) with the appearances of 3D building model. The yellow and blue lines indicate the true trajectory and position solution of a GNSS receiver embedded in a commercial FCS, respectively.