SlideSLAM: Sparse, Lightweight, Decentralized Metric-Semantic SLAM for Multi-Robot Navigation

Abstract: This paper develops a real-time decentralized metric-semantic Simultaneous Localization and Mapping (SLAM) algorithm framework that enables a heterogeneous robot team to collaboratively construct object-based metric-semantic maps of real-world environments featuring indoor, urban, and forests without relying on GPS. The framework integrates a data-driven front-end for instance segmentation from either RGBD cameras or LiDARs and a custom back-end for optimizing robot trajectories and object landmarks in the map. To allow multiple robots to merge their information, we design semantics-driven place recognition algorithms that leverage the informativeness and viewpoint invariance of the object-level metric-semantic map for inter-robot loop closure detection. A communication module is designed to track each robot's observations and those of other robots whenever communication links are available. Our framework enables real-time decentralized operations onboard robots, allowing them to leverage communication opportunistically. We integrate the proposed framework with the autonomous navigation and exploration systems of three types of aerial and ground robots, conducting extensive experiments in a variety of indoor and outdoor environments. These experiments demonstrate its accuracy in inter-robot localization and object mapping, along with its moderate demands on computation, storage, and communication resources. The framework is open-sourced and is suitable for both single-agent and multi-robot metric-semantic SLAM applications. The project website and code can be found at https://xurobotics.github.io/slideslam/ and https://github.com/XuRobotics/SLIDE_SLAM, respectively.

• The paper presents SlideSLAM, a decentralized, sparse metric-semantic SLAM system designed for efficient multi-robot navigation in GPS-denied, complex 3D environments.
• It achieves high precision with an average inter-robot localization error of 20 cm and 0.2° orientation while maintaining an object mapping F1 score above 0.9.
• The system reduces computational load and communication overhead, enabling robust deployment on SWaP-constrained platforms across diverse settings.

Overview of "SlideSLAM: Sparse, Lightweight, Decentralized Metric-Semantic SLAM for Multi-Robot Navigation"

In the academic paper titled "SlideSLAM: Sparse, Lightweight, Decentralized Metric-Semantic SLAM for Multi-Robot Navigation", the authors introduce SlideSLAM, a real-time decentralized metric-semantic SLAM (Simultaneous Localization and Mapping) system. This system is specifically developed for multi-robot navigation and aims to provide a solution for environments where traditional approaches may struggle due to the complexity and scale of the task. The robots operate in three-dimensional settings that include configurations ranging from indoor spaces to urban and forested areas, without relying on GPS capabilities.

SLAM Representation and Architecture

The proposed system uses a hierarchical metric-semantic representation, integrating high-level semantic maps of objects and voxel maps. The system leverages object-based representations, which are sparse and lightweight, allowing efficient handling of the Simultaneous Localization and Mapping problem. High-level semantic maps include sparse semantic models that greatly facilitate place-recognition and loop closure detection across different robot platforms, including those with varying sensing modalities. The authors have integrated a communication module that manages the observations of each robot and other interacting robots to construct a comprehensive merged map.

Technical Results and Performance Metrics

The empirical results provided in the paper are particularly significant, showcasing SlideSLAM's effectiveness. Notably, the average inter-robot localization error achieved is approximately 20 cm for position and 0.2 degrees for orientation. This level of precision marks a considerable achievement in the domain of multi-robot SLAM. Furthermore, the object mapping F1 score consistently attains values over 0.9, signifying robust performance in recognizing and mapping semantic objects in the environment. Crucially, the communication overhead is kept minimal—a packet size of merely 2-3 megabytes per kilometer trajectory involving up to 1,000 landmarks, ensuring the system's efficiency even under constrained bandwidth conditions.

Practical and Theoretical Implications

The research reveals several practical implications. Firstly, SlideSLAM significantly reduces computational load while maintaining accuracy, making it ideal for Size, Weight, and Power (SWaP) constrained platforms. This development promotes the feasibility of deploying autonomous, heterogeneous robot teams in complex real-world environments without extensive reliance on centralized infrastructure or GPS. Theoretically, the integration of semantics into the SLAM back-end optimization process hints at broader applications of semantic information, potentially transforming object detection, mapping, and navigation tasks in robotics.

Future Directions

Given its promising results, the research suggests several future directions. The developed system can be further enhanced by integrating more sophisticated semantic object detection techniques. Additionally, improving the inter-robot communication strategies and exploring further optimization of resource allocation for extensive environments could expand SlideSLAM's applications. Further research might also consider scaling the system to larger fleet sizes and even more challenging environments, potentially using learning-based methodologies to dynamically adapt the hierarchical semantic representation.

In conclusion, SlideSLAM introduces a meaningful advancement in decentralized multi-robot navigation, demonstrating the potential of combining metric-semantic representations with lightweight, efficient SLAM systems for diverse, GPS-denied settings. This work stands as a robust foundation for exploring more advanced and adaptive SLAM systems in the future.