S-Graphs 2.0 -- A Hierarchical-Semantic Optimization and Loop Closure for SLAM

TL;DR

This work tackles the scalability challenge of SLAM in large multi-floor indoor environments by introducing S-Graphs 2.0, a four-layer hierarchical SLAM graph that explicitly incorporates floor levels. It adds floor segmentation and stairway detection to assign floor identities and leverages three optimization strategies—local, floor-level global, and room-level local—to maintain accuracy while reducing computation. Floor-level loop closures prevent cross-floor false positives, and room-level marginalization minimizes redundant observations within rooms. The approach achieves state-of-the-art accuracy and substantial computation reductions (up to 10x faster on baselines and over 90% faster in multi-floor tests) and demonstrates robust floor separation and loop-closure behavior across floors. This enables real-time, semantically informed SLAM suitable for large indoor spaces, with future work targeting elevator transitions and richer semantic relations.

Abstract

The hierarchical structure of 3D scene graphs shows a high relevance for representations purposes, as it fits common patterns from man-made environments. But, additionally, the semantic and geometric information in such hierarchical representations could be leveraged to speed up the optimization and management of map elements and robot poses. In this direction, we present our work Situational Graphs 2.0 (S-Graphs 2.0), which leverages the hierarchical structure of indoor scenes for efficient data management and optimization. Our algorithm begins by constructing a situational graph that represents the environment into four layers: Keyframes, Walls, Rooms, and Floors. Our first novelty lies in the front-end, which includes a floor detection module capable of identifying stairways and assigning floor-level semantic relations to the underlying layers. Floor-level semantics allows us to propose a floor-based loop closure strategy, that effectively rejects false positive closures that typically appear due to aliasing between different floors of a building. Our second novelty lies in leveraging our representation hierarchy in the optimization. Our proposal consists of: (1) local optimization over a window of recent keyframes and their connected components across the four representation layers, (2) floor-level global optimization, which focuses only on keyframes and their connections within the current floor during loop closures, and (3) room-level local optimization, marginalizing redundant keyframes that share observations within the room, which reduces the computational footprint. We validate our algorithm extensively in different real multi-floor environments. Our approach shows state-of-art-art accuracy metrics in large-scale multi-floor environments, estimating hierarchical representations up to 10x faster, in average, than competing baselines

Figures (9)

• Figure 1: Three-story map generated by our S-Graphs 2.0. Floor levels are color-coded in green, red, and orange, while the estimated robot trajectory is shown in black.
• Figure 2: System Architecture. The inputs to our method are the 3D LiDAR data along with the odometry measurements. Its contains different modules in the front-end modules for generating the four-layered hierarchical graph and organizing it into floor-level. A back-end module which exploits the hierarchy in the graph to apply different optimization strategies.
• Figure 3: Stairway Detection, with the keyframes corresponding to stairways as blue dots, and those corresponding to different floor levels as red and green dots.
• Figure 4: Local Optimization. Orange- and blue-colored keyframes, along with their connected layers, are included in the local optimization, with blue keyframes being fixed. Red keyframes are not incorporated in the optimization, as they are outside the optimization window.
• Figure 5: Floor-level Global Optimization. Floor-1 keyframes with orange colored boxes along with the connected wall, room and floor nodes are included in the floor-level global optimization on detection of a loop closure candidate on floor-1. Floor-0 (lower level) maintains its own independent pose graph structure with keyframes (orange spheres) with connected rooms (pink and green cubes) and walls. These elements are temporarily excluded from the Floor-1 optimization to reduce computational complexity and prevent cross-floor interference.