A fast and high precision multi-robot environment modeling based on M-BFSI: Bidirectional filtering and scene...

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Multi-robot systems require accurate environment modeling for effective collaboration.
• Challenges in simultaneous localization and mapping (SLAM) include feature tracking failures due to large rotations.
• Existing methods may struggle with large-scale, dynamic environments and data association.

Purpose of the Study:

• To design and implement a fast, high-precision multi-robot environment modeling method.
• To address feature tracking failures in multi-robot SLAM caused by large angle rotations.
• To enable efficient scene identification and cooperative mapping across multiple robots.

Main Methods:

• A bidirectional filtering mechanism is integrated to enhance error-matching elimination algorithms.
• A global key frame database using a pre-trained dictionary converts images into bag-of-words vectors for efficient scene retrieval.
• Cooperative SLAM is achieved through similarity score calculation, best image matching, and transformation matrix application.

Main Results:

• The proposed algorithm effectively closes the predicted trajectory of sub-robots, ensuring localization accuracy.
• High-precision collaborative environment modeling was demonstrated through experimental validation.
• The method shows robustness in identifying and utilizing similar scenes across different robots.

Conclusions:

• The developed method provides a robust and accurate solution for multi-robot environment modeling.
• Bidirectional filtering and scene identification significantly improve collaborative SLAM performance.
• This approach facilitates efficient and precise mapping in complex environments for multi-robot systems.