Abstract
An integrated navigation system is a promising solution to improve positioning performance by complementing estimated positioning in each sensor, such as a global positioning system (GPS), an inertial measurement unit (IMU), and an odometer sensor. However, under GPS-disabled environments, such as urban canyons or tunnels where the GPS signals are difficult to receive, the positioning performance of the integrated navigation system decreases. Therefore, deep learning-based integrated navigation systems have been proposed to ensure seamless localization under various positioning conditions. Nevertheless, the conventional deep learning-based systems are applied with a lack of consideration of context features on surface condition, wheel slip, and movement pattern, which are factors causing positioning performance. In this paper, a context-aware integrated navigation system (CAINS) is proposed to ensure seamless localization, especially under GPS-disabled conditions. In the proposed CAINS, two deep learning layers are designed with context-aware and state estimation layers. The context-aware layer extracts vehicle context features from IMU data, while the state estimation layer predicts the GPS position increments by modeling the relationship between context features, velocity, attitude, and position increments. From simulation results, it is confirmed that the positioning accuracy can be significantly improved based on the proposed CAINS when compared with conventional navigation systems.