Multi-sensor localization - Visual Odometry as a low cost proprioceptive sensor

Abstract

Ego-localization is a key issue for most autonomous robots and vehicles. Indeed, the ability to take a proper decision (avoidance, path-finding, etc.) relies on the knowledge of one's particular environment on one hand and on its relative positioning in this environment on the other hand. As such, this issue has been addressed multiple times in the past few years. This work extends a multi-sensor fusion framework in order to take advantage of Visual Odometry (VO), as a low cost proprioceptive sensor with the same result than an expensive INS sensor. In particular, it is shown that VO helps to determine the course of the vehicle and to limit the overall drift of the system with a similar behavior than with a classical but expensive localization filter.