Multi-Sensor Fusion for Wheel-Inertial-Visual Systems Using a Fuzzification-Assisted Iterated Error State Kalman...

Area of Science:

• Robotics
• Sensor Fusion
• Artificial Intelligence

Background:

• Odometry drift is a significant challenge for indoor mobile robots, particularly in unstructured environments.
• Traditional localization methods struggle with dynamic conditions, varying lighting, and complex robot kinematics.

Purpose of the Study:

• To propose a multi-sensor fusion framework for accurate 6-DoF (six degrees-of-freedom) localization of differential-drive indoor mobile robots.
• To enhance robot localization robustness in unstructured and dynamic indoor scenes.

Main Methods:

• Developed a Wheel-Inertial-Visual Odometry (WIVO) framework incorporating a Fuzzy Inference System (FIS).
• Integrated FIS with an Iterated Error State Kalman Filter (IESKF) to adaptively adjust noise covariance matrices.
• Optimized fuzzy inference rule parameters for dynamic noise prediction.

Main Results:

• The proposed FIS-IESKF fusion method significantly improves localization accuracy compared to traditional approaches.
• Demonstrated enhanced system robustness for differential-drive robots in dynamic environments and movements.
• Successfully addressed limitations of fixed covariance matrices in Kalman filtering for odometry.

Conclusions:

• The multi-sensor fusion approach using FIS-IESKF is effective for precise and robust 6-DoF indoor mobile robot localization.
• Adaptive noise handling via FIS improves performance in challenging, unstructured, and dynamic environments.
• This framework offers a promising solution for reliable odometry in real-world robotic applications.