Stereo SLAM for smart glasses

Abstract

Developed a lightweight stereo visual-inertial SLAM system for mobile XR platforms with these features:

• Real-time tracking at 90Hz on resource-limited devices.
• Accurate pose estimation with <1cm drift over 50m.
• Support for multiple operating systems: Android, Windows, Linux, macOS, ROS.
• Modular architecture for integration with smart glasses and XR systems.

Problem

• Limited CPU/GPU power for heavy SLAM computations
• Need for high-frequency (90Hz) real-time tracking
• Sensor fusion complexities with stereo cameras and IMUs
• Cross-platform deployment difficulties

Contribution

• Designed and implemented a modular, multithreaded SLAM pipeline optimized for ARM-based mobile SoCs
• Built a sensor fusion system integrating stereo vision and IMU data for stable 6DoF pose estimation
• Tuned performance to run efficiently on embedded and mobileplatforms (Galaxy S8–S22, Realsense D435i)
• Developed cross-platform abstractions supporting Android, Linux, macOS, Windows, and ROS

Result

• Achieved <1cm pose drift over 50 metersin indoor/outdoor tests
• Enabled 90Hz real-time SLAM on various mobile devices including Galaxy S8–S22
• Successfully integrated with tethered smart glasses for XR applications