Continuous Positioning with Recurrent Auto-Regressive Neural Network for Unmanned Surface Vehicles in GPS Outages

As the vital operation information of unmanned surface vehicles, the positioning data are usually measured with GPS/INS (Global Position System/Inertial Navigation System) which faces measurement loss and calculation failure during GPS outages in a complex environment. A continuous positioning method is proposed based on an improved neural network with the available sensor data. Firstly, the continuous positioning framework is built to synthesize the traditional GPS/INS coupling mode with the novel estimation method of the improved neural network. Secondly, a reconstructed model of the recurrent auto-regressive neural network is proposed with dual-loop structures, which can excavate the time-series features and the nonlinear relation in multiple sensor measurements. Thirdly, the continuous inertial positioning algorithm is designed based on the novel network, in which the alignment of measurement data is studied to form the augmented inputs. Finally, different experiments are designed and conducted to verify the method, including the outage performance, estimation duration, and model comparison. The results show that positioning estimation precision is relatively high, and the estimation duration reaches an acceptable degree. The proposed method is feasible and effective for positioning in GPS outages.