Human-Inspired Long-Term Indoor Localization in Human-Oriented Environment

TL;DR

An overview of the past research on long-term localization and mapping, exploiting geometric priors such as floor plans and integrating textual and semantic information is presented.

Abstract

Lifelong localization is crucial for enabling the autonomy of service robots. In this paper, we present an overview of our past research on long-term localization and mapping, exploiting geometric priors such as floor plans and integrating textual and semantic information. Our approach was validated on challenging sequences spanning over many months, and we released open source implementations.

Figures (2)

• Figure 1: A 2D abstract semantic map enriching a floor plan with semantic information, used for long term localization. Different box colors indicate different object classes.
• Figure 2: (a) The text likelihood maps (b) Particle distribution prior to text spotting, with multi-modal distribution (c) Particle injection based on the text likelihood map after a textual cue was detected.