MCN-SLAM: Multi-Agent Collaborative Neural SLAM with Hybrid Implicit Neural Scene Representation (2506.18678v1)

Abstract: Neural implicit scene representations have recently shown promising results in dense visual SLAM. However, existing implicit SLAM algorithms are constrained to single-agent scenarios, and fall difficulties in large-scale scenes and long sequences. Existing NeRF-based multi-agent SLAM frameworks cannot meet the constraints of communication bandwidth. To this end, we propose the first distributed multi-agent collaborative neural SLAM framework with hybrid scene representation, distributed camera tracking, intra-to-inter loop closure, and online distillation for multiple submap fusion. A novel triplane-grid joint scene representation method is proposed to improve scene reconstruction. A novel intra-to-inter loop closure method is designed to achieve local (single-agent) and global (multi-agent) consistency. We also design a novel online distillation method to fuse the information of different submaps to achieve global consistency. Furthermore, to the best of our knowledge, there is no real-world dataset for NeRF-based/GS-based SLAM that provides both continuous-time trajectories groundtruth and high-accuracy 3D meshes groundtruth. To this end, we propose the first real-world Dense slam (DES) dataset covering both single-agent and multi-agent scenarios, ranging from small rooms to large-scale outdoor scenes, with high-accuracy ground truth for both 3D mesh and continuous-time camera trajectory. This dataset can advance the development of the research in both SLAM, 3D reconstruction, and visual foundation model. Experiments on various datasets demonstrate the superiority of the proposed method in both mapping, tracking, and communication. The dataset and code will open-source on https://github.com/dtc111111/mcnslam.

• The paper introduces MCN-SLAM, a multi-agent collaborative neural SLAM framework leveraging a hybrid implicit neural scene representation.
• It employs a triplane-grid hybrid scene representation and online distillation for efficient submap fusion and improved accuracy and scalability.
• The framework demonstrates enhanced mapping and tracking performance compared to existing methods and includes the introduction of the Dense SLAM Dataset (DES).

Overview of MCN-SLAM: Multi-Agent Collaborative Neural SLAM

The paper "MCN-SLAM: Multi-Agent Collaborative Neural SLAM with Hybrid Implicit Neural Scene Representation" presents an innovative approach to multi-agent collaborative simultaneous localization and mapping (SLAM) designed to overcome challenges posed by existing methodologies. Leveraging neural implicit scene representations, the authors propose a framework that addresses the limitations of single-agent systems and extends current NeRF-based multi-agent frameworks by improving communication efficiency and scalability in large environments.

Key Contributions

1. Hybrid Scene Representation: The paper introduces a novel triplane-grid joint scene representation method that effectively captures both low-frequency structural features and high-frequency textures. This hybrid approach facilitates faster convergence and hole filling in unobserved areas, significantly enhancing the accuracy and completeness of the reconstructed scenes.
2. Intra-to-Inter Loop Closure: The authors propose a method to ensure both local and global consistency. This involves intra-loop closure within a single agent's keyframe graph and inter-loop closure that registers and optimizes poses across multiple agents, minimizing cumulative errors and ensuring global map consistency.
3. Online Distillation for Submap Fusion: The technique implemented achieves seamless integration of submaps obtained by different agents. By employing peer-to-peer online distillation involving neural network parameter exchange, the framework efficiently fuses information across agents to construct a consistent global map.
4. DES Dataset Introduction: A notable contribution is the introduction of the Dense SLAM Dataset (DES), which fills the gap in NeRF-based and GS-based SLAM research by providing comprehensive real-world data with high-accuracy 3D mesh and continuous camera trajectory ground truth for both single and multi-agent scenarios.

Numerical Results and Experimental Setup

The experiments conducted on various datasets demonstrate the superiority of MCN-SLAM in mapping, tracking, and communication compared to existing methods. Notably, the framework achieves improved surface reconstruction quality, reduced pose estimation errors, and efficient bandwidth usage.

Implications and Future Directions

The development of this distributed multi-agent SLAM framework has meaningful implications for industries where large-scale situational awareness is required, such as autonomous driving, robotics, and surveillance. The introduction of the DES dataset can propel the field forward by providing benchmarks for future research.

Looking ahead, the combination of neural implicit representations with multi-agent SLAM frameworks opens avenues for exploring more efficient scene encoding techniques, optimizing peer-to-peer communication protocols, and scaling the framework to even larger and more complex environments.

Conclusion

Overall, the MCN-SLAM framework represents a significant step toward advancing collaborative SLAM systems. Its hybrid scene representation, robust loop closure mechanisms, and efficient submap fusion processes collectively enhance the capabilities of multi-agent systems in achieving high accuracy and scalability in dynamic and expansive environments.