Active Localization Strategy for Hypotheses Pruning in Challenging Environments

Robust localization system has proven to be a cornerstone for mobile robot autonomy. Although passive robot localization is a mature field, it still could fail in challenging environments containing symmetries or open spaces. Active localization can fix this issue by allowing the robot to improve po...

Full description

Saved in:
Bibliographic Details
Published in:Journal of intelligent & robotic systems 2022-10, Vol.106 (2), Article 47
Main Authors: Andrade, Federico, Llofriu, Martín, Marzoa Tanco, Mercedes, Trinidad Barnech, Guillermo, Tejera, Gonzalo
Format: Article
Language:English
Subjects:
Algorithms
Artificial Intelligence
Clustering
Control
Electrical Engineering
Engineering
Hypotheses
Indoor environments
Localization
Mechanical Engineering
Mechatronics
Open spaces
Pose estimation
Regular Paper
Robotics
Robots
Topical Collection in Robotics in South America (S)
Tracking problem
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Robust localization system has proven to be a cornerstone for mobile robot autonomy. Although passive robot localization is a mature field, it still could fail in challenging environments containing symmetries or open spaces. Active localization can fix this issue by allowing the robot to improve pose estimation by choosing specific actions. We propose an active localization strategy for the indoor position tracking problem in challenging environments. The proposed active localization is performed in three steps: (i) cluster the particle cloud with Spectral Clustering (or Kmeans + + ) algorithm, (ii) search and select the most informative point in a reduced search space, and (iii) execute rotational actions in order to sense the selected point. Hence, a significant number of wrong hypotheses are pruned. We also introduce a novel study that considers evaluates points in spatial neighborhoods all at once, instead of evaluating each cell independently. Simulated experiments show an improvement in robot pose estimation using the proposed strategy. Real-world validation in symmetric and open office-like environment is also presented.
ISSN:0921-0296
1573-0409
DOI:10.1007/s10846-022-01748-4