Loop-Box: Multiagent Direct SLAM Triggered by Single Loop Closure for Large-Scale Mapping.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Traditional simultaneous localization and mapping (SLAM) often relies on complex optimization and multiple sensors.
• Existing multiagent systems frequently depend on odometry from additional sensors, increasing system complexity.
• Current methods are often tightly coupled with specific sensors and extensive computer vision algorithms.

Purpose of the Study:

• To present a versatile multiagent framework for real-time, large-scale 3-D reconstruction.
• To address challenges in multiagent 3-D mapping across diverse camera systems.
• To enable accurate localization and global mapping using only monocular cameras.

Main Methods:

• Development of a generic framework for multiagent 3-D mapping.
• Implementation of active agent localization post-loop closure.
• Utilizing monocular cameras for localization and mapping.
• Calculating optimal scale differences and relative pose transformations between maps.

Main Results:

• Real-time multiagent large-scale localization and 3-D global mapping achieved.
• The framework effectively localizes agents after initial loop closure.
• Accurate relative pose estimation for global mapping based on initial matching.
• Successful application with different camera systems.

Conclusions:

• The proposed framework offers a robust solution for real-time multiagent 3-D mapping.
• Monocular cameras are sufficient for achieving high-quality large-scale localization and mapping.
• The system provides accurate scale and pose estimation for integrated multiagent maps.