Robust Tightly Coupled Pose Measurement Based on Multi-Sensor Fusion in Mobile Robot System

Area of Science:

• Robotics
• Computer Vision
• Sensor Fusion

Background:

• Simultaneous Localization and Mapping (SLAM) is crucial for mobile robots.
• Monocular visual-inertial SLAM faces scale ambiguity and robustness challenges.
• Wheel speed sensors can enhance SLAM robustness and resolve scale issues for ground robots.

Purpose of the Study:

• To propose a tightly-coupled, multi-sensor fusion SLAM algorithm integrating monocular vision, inertia, and wheel speed measurements.
• To improve the accuracy, robustness, and localization capabilities of SLAM systems for ground mobile robots.
• To introduce novel wheel odometer pre-integration and state initialization methods for enhanced performance.

Main Methods:

• A tightly-coupled nonlinear optimization approach is employed for state estimation.
• Loop detection and back-end optimization are utilized to ensure global consistency.
• A wheel odometer pre-integration algorithm combines chassis and IMU angular velocities.
• State initialization leverages wheel odometer and IMU data for stationary and moving states.

Main Results:

• The proposed algorithm achieved high accuracy with 0.28% cumulative error over 812m (loopback optimization disabled).
• Demonstrated robustness in various scenarios, including sensor loss (e.g., visual loss).
• Outperformed monocular visual-inertial SLAM and traditional wheel odometers in accuracy and robustness.

Conclusions:

• The multi-sensor fusion SLAM algorithm offers superior accuracy and robustness for ground mobile robots.
• The novel pre-integration and initialization methods enhance SLAM performance significantly.
• The system provides effective localization even during sensor failures, ensuring reliable navigation.