An Enhanced Hybrid Visual-Inertial Odometry System for Indoor Mobile Robot

Area of Science:

• Robotics
• Sensor Fusion
• State Estimation

Background:

• Accurate mobile robot localization is crucial for autonomous systems.
• Multi-sensor fusion, particularly Visual-Inertial Odometry (VIO), enhances localization accuracy and robustness over single-sensor systems.
• Existing VIO methods like MSCKF often underutilize IMU data after initial state prediction.

Purpose of the Study:

• To improve the positioning accuracy and robustness of mobile robots.
• To enhance the Multi-State Constraint Kalman Filter (MSCKF) framework by incorporating IMU pre-integration results as observational information.
• To optimize sensor fusion weighting and leverage wheel odometer data for enhanced localization.

Main Methods:

• Proposed a novel framework integrating IMU pre-integration results into the MSCKF framework as observation information.
• Employed Helmert variance component estimation (HVCE) to dynamically adjust weights between visual features and IMU pre-integration.
• Utilized mobile robot wheel odometer data for zero-speed detection and updates to refine pre-integration estimates.

Main Results:

• The proposed algorithm demonstrated superior positioning accuracy compared to existing mainstream VIO algorithms.
• Real-time performance was maintained, ensuring the system's practicality for mobile robot applications.
• Experiments in simulation (Gazebo), public datasets, and real-world scenarios validated the algorithm's effectiveness.

Conclusions:

• The novel MSCKF framework with IMU pre-integration significantly improves mobile robot localization accuracy.
• The integration of HVCE and wheel odometer data further enhances robustness and precision.
• The developed algorithm offers a promising solution for accurate and real-time mobile robot navigation.