Open-World Distributed Robot Self-Localization with Transferable Visual Vocabulary and Both Absolute and Relative Features

TL;DR

A new self-localization framework designed for open-world distributed robot systems that maintains state-of-the-art performance while offering two key advantages: (1) it employs an unsupervised visual vocabulary model that maps to multimodal, lightweight, and transferable visual features, and (2) the visual vocabulary itself is a lightweight and communication-friendly model.

Abstract

Visual robot self-localization is a fundamental problem in visual robot navigation and has been studied across various problem settings, including monocular and sequential localization. However, many existing studies focus primarily on single-robot scenarios, with limited exploration into general settings involving diverse robots connected through wireless networks with constrained communication capacities, such as open-world distributed robot systems. In particular, issues related to the transfer and sharing of key knowledge, such as visual descriptions and visual vocabulary, between robots have been largely neglected. This work introduces a new self-localization framework designed for open-world distributed robot systems that maintains state-of-the-art performance while offering two key advantages: (1) it employs an unsupervised visual vocabulary model that maps to multimodal, lightweight, and transferable visual features, and (2) the visual vocabulary itself is a lightweight and communication-friendly model. Although the primary focus is on encoding monocular view images, the framework can be easily extended to sequential localization applications. By utilizing complementary similarity-preserving features -- both absolute and relative -- the framework meets the requirements for being unsupervised, multimodal, lightweight, and transferable. All features are learned and recognized using a lightweight graph neural network and scene graph. The effectiveness of the proposed method is validated in both passive and active self-localization scenarios.

Figures (5)

• Figure 1: Visual place recognition (VPR) from a novel highly compressive sequential semantic scene graph (S3G) is considered. To address the information lost in dimension reduction, the appearance/spatial image information is mapped to two different features, absolute (node) and relative (edge) features, which complement each other. Additionally, a new task of viewpoint planning of the query S3G is enabled by the trained VPR, to further improve the VPR performance.
• Figure 2: Importance of edges. In highly compressive applications, a naive strategy of using only absolute features (i.e., nodes) suffer from information loss during dimension reduction. To address this issue, we exploit the edges as relative features that complement the absolute features.
• Figure 3: Experimental environments. The trajectories of the four datasets, "2012/1/22," "2012/3/31," "2012/8/4," and "2012/11/17," used in our experiments are visualized in green, purple, blue, and light-blue curves, respectively.
• Figure 4: S2G examples. Top: The input image. Bottom: S2G overlaid on the semantic label image.
• Figure 5: NBV planning results. In each figure, the bottom and top panels show the view image before and after planned movements, respectively.