DGOMapping: Real-Time Multi-Agent Mapping Based on 4D Gaussian Splatting.

Area of Science:

• Robotics and Computer Vision
• Simultaneous Localization and Mapping (SLAM)

Background:

• 3D Gaussian Splatting (3DGS) excels at detailed scene reconstruction but struggles with dynamic environments and multi-agent scalability.
• Existing multi-agent mapping methods lack robust handling of temporal asynchrony and dynamic inconsistencies, leading to conflicting updates.

Purpose of the Study:

• To introduce DGOMapping, an online system for robust multi-agent dynamic perceptual mapping.
• To address limitations in current 3DGS-based mapping for complex, real-world scenarios.

Main Methods:

• Utilizes an uncertainty-coupled 4D Gaussian Splatting (4DGS) scene representation for enhanced detail and temporal consistency.
• Implements a collaborative interaction mechanism through Gaussian perception-score exchange for robust multi-agent coordination.

Main Results:

• Demonstrates effective suppression of dynamic interference in real-world datasets.
• Achieves state-of-the-art performance in both tracking and reconstruction through multi-agent collaboration.
• Validates real-time 4DGS construction and long-term map memory adjustment capabilities.

Conclusions:

• DGOMapping provides a practical, sensing-oriented solution for collaborative perception and real-time dynamic environment mapping.
• The system's ability to manage uncertainty and asynchronous observations improves adaptability and real-world applicability in multi-agent systems.