Multi-Sensor Fusion for Mono and Multi-Vehicle Localization using Bayesian Network

Abstract

Outdoor mobile robotised vehicles currently hold the attention of many researchers because they can bring solutions to many applications related to transport of passengers in urban environments. An example of robotised vehicle is the CyCab (Bom et al., 2005). Transport applications which can combine mono or multi vehicle operating mode are most efficient. For autonomous navigation application, the vehicle needs to know its position accurately (Dissanayake et al. 2001); (Thrun et al. 2001) and if possible on the road network. In this work, we propose to use the digital road map database for the geo-localization of the vehicle. The localization of a vehicle using or in respect to or on a road map is treated by several ways in the last ten years. This relatively recent research theme is known also as the map-matching or road-matching problem. It can be interesting to localize a vehicle using a road map because it can useful to recover the attributes associated with these data bases. Examples of attributes are the width of the road, the presence of landmarks for accurate localization, authorized maximum speed for advanced driver assistance system application etc. Unfortunately, the use of the road map to improve the localization is not a simple task. There are always errors on the estimate of the position and because the map can represents a deformed sight of the world. Outdoor positioning systems often rely on GPS, because of its affordability and convenience. However, GPS suffers from satellite masks occurring in urban environments, under bridges, tunnels or in forests. GPS appears then as an intermittently-available positioning system that needs to be backed up by a dead-reckoning system (Zhao, 1997); (Abbott & Powell, 1999); (EL Najjar & Bonnifait, 2003). In this work, the proposed method of multi-sensors fusion for mono-vehicle localization is based on the use of encoders positioned at the rear wheel of the vehicle. We use these sensors to measure elementary rotations of the wheels and to estimate the displacement of the vehicle. Thus, a dead-reckoned estimated pose is obtained by integrating the elementary rotations of the wheels using a differential odometric model. The multisensor fusion of GPS and odometry is performed by a Bayesian Network (BN). Afterwards, we extend the multi-sensor fusion method proposed in this work for mono-vehicle localization to be used for the localization of several vehicles moving in the same environment. We suppose in this extension that the vehicles evolve in a train configuration. In the literature, we found two ways to make moving a train of vehicles. The first one