An Audio-Based SLAM for Indoor Environments: A Robotic Mixed Reality Presentation

Area of Science:

• Robotics
• Computer Vision
• Acoustics

Background:

• Traditional Simultaneous Localization and Mapping (SLAM) often relies on visual landmarks, which can be unreliable in certain environments.
• Audio-based SLAM methods typically require multiple sound sources or sensor fusion for accurate localization.
• Existing audio SLAM systems face challenges with sparse or indiscernible audio landmarks.

Purpose of the Study:

• To develop a novel audio-based SLAM system using virtual landmarks for enhanced robot navigation and mapping.
• To enable autonomous robots to localize and map environments using a single sound source and microphone arrays.
• To address the limitations of conventional audio-based SLAM in environments with poor or missing audio landmarks.

Main Methods:

• Implemented an audio-based virtual landmark system (HoloSLAM) utilizing a single sound source and microphone arrays.
• Integrated Robotic Mixed Reality (MR) with Microsoft HoloLens to superimpose virtual audio landmarks.
• Employed an active speaker detection method for sound source localization using only audio data.

Main Results:

• The HoloSLAM system demonstrated accurate speaker direction estimation and robot self-localization.
• The approach successfully mapped indoor environments without relying on multiple audio sources or sensor fusion.
• Real-time experiments validated the system's precision, efficiency, and comprehensive mapping capabilities, surpassing traditional methods.

Conclusions:

• The proposed audio-based virtual landmark HoloSLAM offers a robust solution for robot navigation and mapping in challenging acoustic environments.
• This method effectively overcomes the limitations of conventional audio SLAM by utilizing virtual landmarks and mixed reality.
• The system provides accurate localization and detailed environmental mapping, paving the way for more autonomous robotic applications.