Neural network-based multiple robot simultaneous localization and mapping

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision

Background:

• Simultaneous Localization and Mapping (SLAM) is crucial for autonomous robot navigation.
• Extending single-robot SLAM to multi-robot systems presents significant challenges in map fusion and coordination.

Purpose of the Study:

• To develop a decentralized platform for multi-robot simultaneous localization and mapping (SLAM).
• To propose a novel occupancy grid map fusion algorithm for enhanced multi-robot coordination.

Main Methods:

• Each robot employs view-based SLAM with an extended Kalman filter, fusing encoder and laser ranger data.
• A multistep map fusion algorithm is introduced, including neural network-based clustering for map learning.
• Relative orientation and translation are extracted using norm histogram cross-correlation, Radon transform, and matching norm vectors.

Main Results:

• The proposed self-organizing map-based clustering effectively learns map features for unsupervised, on-the-fly map fusion.
• The system successfully extracts relative poses between robots for accurate map merging.
• Experimental results in a real environment demonstrate the effectiveness of the multi-robot SLAM solution.

Conclusions:

• The developed decentralized platform enables efficient multi-robot SLAM.
• The novel map fusion algorithm, leveraging unsupervised learning, significantly improves multi-robot localization and mapping accuracy.
• The approach offers a robust solution for complex robotic navigation tasks.