TACS-Graphs: Traversability-Aware Consistent Scene Graphs for Ground Robot Localization and Mapping

TL;DR

This work tackles segmentation inconsistencies in 3D indoor scene graphs by replacing voxel-based free-space representations with a traversability-aware framework, TACS-Graphs. It introduces traversable graph extraction from LiDAR data (via B-TMS and BGK inference), wall- and free-space-based room segmentation, and a room-merging module to ensure topological consistency. A novel loop-closure detector, CoSG-LCD, leverages consistent room structure to trigger targeted loop closures, improving pose graph optimization and localization accuracy. Extensive experiments on TIERS, S-Graphs, and real-world datasets show improved scene-graph consistency (higher DCS, lower room-std) and enhanced PGO performance, validating the practical impact for robust ground robot SLAM in complex environments. The approach advances semantic mapping by aligning room boundaries with traversability, reducing misclassifications in open spaces and enhancing loop-closure efficiency.

Abstract

Scene graphs have emerged as a powerful tool for robots, providing a structured representation of spatial and semantic relationships for advanced task planning. Despite their potential, conventional 3D indoor scene graphs face critical limitations, particularly under- and over-segmentation of room layers in structurally complex environments. Under-segmentation misclassifies non-traversable areas as part of a room, often in open spaces, while over-segmentation fragments a single room into overlapping segments in complex environments. These issues stem from naive voxel-based map representations that rely solely on geometric proximity, disregarding the structural constraints of traversable spaces and resulting in inconsistent room layers within scene graphs. To the best of our knowledge, this work is the first to tackle segmentation inconsistency as a challenge and address it with Traversability-Aware Consistent Scene Graphs (TACS-Graphs), a novel framework that integrates ground robot traversability with room segmentation. By leveraging traversability as a key factor in defining room boundaries, the proposed method achieves a more semantically meaningful and topologically coherent segmentation, effectively mitigating the inaccuracies of voxel-based scene graph approaches in complex environments. Furthermore, the enhanced segmentation consistency improves loop closure detection efficiency in the proposed Consistent Scene Graph-leveraging Loop Closure Detection (CoSG-LCD) leading to higher pose estimation accuracy. Experimental results confirm that the proposed approach outperforms state-of-the-art methods in terms of scene graph consistency and pose graph optimization performance.

Figures (9)

• Figure 1: An overview of TACS-Graphs, the proposed traversability-aware scene graph generation framework for ground robots. (a) A real-time traversable graph is extracted using B-TMS oh2024btms. (b) Room segmentation is performed using traversable graphs and walls, resulting in consistent scene graphs. (c) This consistency enhances CoSG-LCD module, a loop closure detection method that recognizes revisited rooms.
• Figure 2: An illustration of free-space clustering and room segmentation using traversable graphs. (a) Process of obtaining $\mathcal{T}_k^{\text{inside}}$ at time $t=k$. (b) An example where $\mathcal{T}^{\text{room}}$ is incrementally formed. Once no new $\mathcal{T}^{\text{inside}}$ is detected at $t=4$ as the robot exits a room, the accumulated $\mathcal{T}^{\text{room}}$ from $t=0$ to $t=3$ is used to define room walls.
• Figure 3: An example of the proposed room merging module. (a) Before merging, two y-direction corridors $\mathcal{C}_1$ and $\mathcal{C}_2$ are distinct, with overlapping traversable graphs and a small center distance $d(x)$. (b) After optimizing keyframes and the center of $\mathcal{C}_2$ to make the wall-to-wall edge converge to zero, they merge, updating $\mathcal{C}_1$ as a unified room.
• Figure 4: Overview of the proposed CoSG-LCD loop closure detection framework. (a) Loop closure detection with a coarse-to-fine strategy. If no keyframe pair exceeds $S_{\text{coarse}}$, a relaxed threshold $S_{\text{fine}}$ is applied to ensure robustness against pose estimation errors. (b) For large rooms, partitions of length $L_{\text{part}}$ are created, and the best keyframe pair in each partition is selected for loop closure if its match score exceeds $S_{\text{coarse}}$.
• Figure 5: Layouts of four TIERS indoor datasets tiers2023benchmark. From left to right: Indoor07, Indoor09, Indoor10, and Indoor11.