Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Platform Teach-and-Repeat Navigation by Visual Place Recognition Based on Deep-Learned Local Features

Václav Truhlařík, Tomáš Pivoňka, Michal Kasarda, Libor Přeučil

TL;DR

This paper tackles robust appearance-based teach-and-repeat navigation in variable environments by replacing classic VPR and shift estimation components with SSM-VPR2 and D2-Net-based horizontal shift computation. The proposed system, SSM-Nav2, supports multi-platform operation including UAVs and uses a particle-filter fused with odometry for localization along a taught trajectory. A new public dataset specifically designed for horizontal shift estimation and VPR evaluation is introduced and used to benchmark against state-of-the-art baselines, showing improved accuracy and robustness in indoors, outdoors, and day-night conditions. The work demonstrates practical impact for long-term autonomous navigation with minimal reliance on precise odometry.

Abstract

Uniform and variable environments still remain a challenge for stable visual localization and mapping in mobile robot navigation. One of the possible approaches suitable for such environments is appearance-based teach-and-repeat navigation, relying on simplified localization and reactive robot motion control - all without a need for standard mapping. This work brings an innovative solution to such a system based on visual place recognition techniques. Here, the major contributions stand in the employment of a new visual place recognition technique, a novel horizontal shift computation approach, and a multi-platform system design for applications across various types of mobile robots. Secondly, a new public dataset for experimental testing of appearance-based navigation methods is introduced. Moreover, the work also provides real-world experimental testing and performance comparison of the introduced navigation system against other state-of-the-art methods. The results confirm that the new system outperforms existing methods in several testing scenarios, is capable of operation indoors and outdoors, and exhibits robustness to day and night scene variations.

Multi-Platform Teach-and-Repeat Navigation by Visual Place Recognition Based on Deep-Learned Local Features

TL;DR

This paper tackles robust appearance-based teach-and-repeat navigation in variable environments by replacing classic VPR and shift estimation components with SSM-VPR2 and D2-Net-based horizontal shift computation. The proposed system, SSM-Nav2, supports multi-platform operation including UAVs and uses a particle-filter fused with odometry for localization along a taught trajectory. A new public dataset specifically designed for horizontal shift estimation and VPR evaluation is introduced and used to benchmark against state-of-the-art baselines, showing improved accuracy and robustness in indoors, outdoors, and day-night conditions. The work demonstrates practical impact for long-term autonomous navigation with minimal reliance on precise odometry.

Abstract

Uniform and variable environments still remain a challenge for stable visual localization and mapping in mobile robot navigation. One of the possible approaches suitable for such environments is appearance-based teach-and-repeat navigation, relying on simplified localization and reactive robot motion control - all without a need for standard mapping. This work brings an innovative solution to such a system based on visual place recognition techniques. Here, the major contributions stand in the employment of a new visual place recognition technique, a novel horizontal shift computation approach, and a multi-platform system design for applications across various types of mobile robots. Secondly, a new public dataset for experimental testing of appearance-based navigation methods is introduced. Moreover, the work also provides real-world experimental testing and performance comparison of the introduced navigation system against other state-of-the-art methods. The results confirm that the new system outperforms existing methods in several testing scenarios, is capable of operation indoors and outdoors, and exhibits robustness to day and night scene variations.

Paper Structure

This paper contains 20 sections, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. SSM-Nav
  4. Other Appearance-Based Systems
  5. Position-Based Systems
  6. Teach-and-Repeat Navigation System
  7. Teaching Phase
  8. Visual Place Recognition
  9. Position Filtering
  10. Horizontal Shift Computation
  11. System for Aerial Vehicles
  12. Dataset for Testing Appearance-Based Teach-and-Repeat Methods
  13. Dataset Description
  14. Metrics
  15. Experiments
  16. ...and 5 more sections

Figures (5)

  • Figure 1: Husky A200 mobile robot and DJI Ryze Tello drone used for experimentation
  • Figure 2: Structure of the introduced T&R system SSM-Nav2
  • Figure 3: Example images of new dataset for shift computation testing
  • Figure 4: Basic (left) and Advanced (right) trajectories driven by SSM-Nav2
  • Figure 5: Example images from external (left) and onboard (right) cameras for day-time reference trajectory and switched-off lights during night-time