Non-linear Variational Bayes Multiple Model for Positioning in Highway Tunnel

This paper focuses on network-based precise positioning for autonomous navigation, in particular in Global Navigation Satellite System (GNSS) denied scenarios, such as highway tunnels. Positioning in such scenarios is harsh due to the multipath propagation and the complex dynamics the vehicle can enact during navigation. A standard solution is to use Bayesian filters with an Interacting Multiple Model (IMM) to handle the heterogeneous motion patterns followed by the vehicle along its route. In this paper, we propose an alternative to the IMM approach, based on Variational Inference (VI), namely Nonlinear Variational Bayes Multiple Model (N-VBMM), to deal with both multiple vehicle dynamics and non-linearities of wireless measurement models. We present tests on real Ultra-Wide Band data recorded in an experimental campaign performed in a highway tunnel, showing that the proposed approach outperforms conventional IMM.