Situationally-aware Path Planning Exploiting 3D Scene Graphs

Saad Ejaz, Marco Giberna, Muhammad Shaheer, Jose Andres Millan-Romera, Ali Tourani · 2025

Read Situationally-aware Path Planning Exploiting 3D Scene Graphs by Saad Ejaz, Marco Giberna, Muhammad Shaheer, Jose Andres Millan-Romera, Ali Tourani (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-06283.txt.

Introduction

Saad Ejaz and a team of co-authors plan robot paths with the help of three dimensional scene graphs, in a paper from 2025. Scene graphs hold both measured layout and meaning for rooms and objects, yet planners rarely use that rich form well.

Their planner, S-Path, first searches a meaning graph to sketch a high level route a person can read at a glance. That sketch marks the zones that matter, so the hard geometry splits into small parts solved side by side. If a route fails, past part answers feed the next try. Tests cut plan time by a large factor.

Catalog record

CREATORSaad Ejaz, Marco Giberna, Muhammad Shaheer, Jose Andres Millan-Romera, Ali Tourani
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE41 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-06283

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Situationally-aware Path Planning Exploiting 3D Scene Graphs Thanks: Authors are with the Automation and Robotics Research Group, Interdisciplinary Centre for Security, Reliability and Trust, University of Luxembourg. Holger Voos is also associated with the Faculty of Science, Technology and Medicine, University of Luxembourg, Luxembourg. {saad.ejaz, marco.giberna, muhammad.shaheer, jose.millan, ali.tourani, p.kremer, holger.voos, joseluis.sanchezlopez}@uni.lu Thanks: * This work was funded, in whole or in part, by the Fonds National de la Recherche of Luxembourg (FNR), DEUS Project (Ref. C22/IS/17387634/DEUS), INVISIMARK project (Ref. DEFENCE22/IS/17800397/INVISIMARK), and RoboSAUR project (Ref. 17097684/RoboSAUR). In addition, it was partially funded by the Institute of Advanced Studies (IAS) of the University of Luxembourg - “Audacity” grant (project TRANSCEND - 2021), and by SATORI project (SATORI/8_euROBIN). Thanks: For the purpose of Open Access, the author has applied a CC-BY-4.0 public copyright license to any Author Accepted Manuscript version arising from this submission. Saad Ejaz, Marco Giberna, Muhammad Shaheer, Jose Andres Millan-Romera, Ali Tourani Abstract 3D Scene Graphs integrate both metric and semantic information, yet their structure remains underexploited for improving path planning efficiency and interpretability. In this work, we present S-Path, a Situationally-aware Path planner that leverages the metric-semantic structure of indoor 3D Scene Graphs to significantly enhance planning efficiency. S-Path follows a two-stage process: it first performs a search over a semantic graph derived from the scene graph to yield a human-understandable high-level path. This also identifies relevant regions for planning, which later allows the decomposition of the problem into smaller, independent subproblems that can be solved in parallel. We also introduce a replanning mechanism that, in the event of an infeasible path, reuses information from previously solved subproblems to update semantic heuristics and prioritize re-use to further improve the efficiency of future planning attempts. Extensive experiments on both real-world and simulated environments show that S-Path achieves average reductions of 5.7x in planning time while maintaining comparable path optimality to classical sampling-based planners, and surpassing them in complex scenarios, making it an efficient and interpretable path planner for environments represented by indoor 3D scene graphs. I Introduction Traditional path planners for indoor robot navigation rely on dense sampling of the configuration space (C-space) to find feasible paths, but their computational cost grows rapidly with its size and complexity, limiting efficiency and usability. To mitigate this, some planners [ 1 , 2 ] incorporate heuristics to sample intelligently , but these heuristics often perform poorly in indoor environments characterized by narrow passageways and limited lines of sight. Moreover, such planners typically lack semantic understanding, making integration with human-intuitive reasoning and commands challenging. To address these limitations, several approaches have incorporated heuristics based on semantic information, such as objects [ 3 , 4 ] and terrain types [ 5 , 6 , 7 ] into the planning process to improve both efficiency and interpretability. However, these methods rely on non-heirarchical, low-level representations, which suffer from scalability and ambiguity issues when applied to large and structurally complex environments. 3D scene graphs [ 8 ] are emerging as efficient data structures that encode spatial relationships between semantic entities in a hierarchical manner, thereby enabling effective organization of complex spaces [ 9 , 10 ] . Fig. 1 : S-Path augments sampling-based planners with high-level metric-semantic scene understanding to restrict sampling to only the relevant regions (green contours), thereby reducing planning time in indoor environments. Recent research has explored the use of 3D scene graphs for task and motion planning [ 11 , 12 , 13 ] ; however, they focus solely on plan feasibility, operate predominantly within the task planning domain, and fail to leverage the potential of high-level semantics (e.g., rooms , doorways , floors ) to improve path planning efficiency. To address this gap, we introduce S-Path, a Situationally-aware Path planner that takes advantage of semantic information from indoor scene graphs to significantly improve planning efficiency while providing human-interpretable path planning. By exploiting the structure of the scene graph, S-Path decomposes the global planning problem into smaller, independent subproblems by performing a search on the high-level semantic layers of the scene graph. These subproblems can then be solved in parallel on a geometric level using classical sampling-based planners, which operate significantly faster on the restricted C-space. Moreover, in the event of environmental changes, S-Path can swiftly replan by focusing only on the affected portions of the problem, while reusing previously solved subproblems to efficiently construct a new path. The main contribution of this paper is a path planning algorithm that: • exploits the hierarchical structure of indoor 3D scene graphs to constrain planning to only the relevant regions of the environment. • decomposes the global planning problem into smaller, independent subproblems, enabling parallel solving, thereby reducing planning time. • includes a resource-efficient replanning strategy that isolates and updates only the affected parts of the plan, thereby reducing replanning latency. Moreover, S-Path is built as a ROS-based framework utilizing the OMPL ( OMPL ) framework [ 14 ] and provides a lightweight interface to exploit the semantics of 3D indoor scene graphs to enhance the efficiency of sampling-based planners. II Related Works II-A Semantics-aware Path Planning Terrain semantics such as the classification of reliable terrain has been employed in path planning to guide UGVs [ 5 ] and UAVs [ 7 , 6 ] . In other works, semantic maps generated using cameras, semantic segmentation, and object detection, have been utilized. Sun et al. [ 15 ] uses multi-level indoor semantic maps encoding obstacles, dynamic entities, and room properties to guide an RRT planner with a path smoother to avoid collision zones. Achat et al. [ 16 ] uses a semantic pointcloud to generate multi-layer semantized maps that guide an A* path planner. Other works, such as [ 17 , 18 ] , utilize semantic information about various objects in the environment to inform path planning for efficient object search. Dharmadhikari et al. [ 3 ] presented a semantics-aware path planner that explores unknown environments, reconstructs meshes, and inspects semantics of interest. Ryll et al. [ 7 ] demonstrated how semantic information improves the perception–estimation–planning loop for high-speed drone flight in urban settings. While the use of semantics in path planning is increasingly common, most approaches rely on flat, low-level representations of the environment semantics, which do not scale effectively in large or complex environments. II-B 3D Scene Graphs for Path Planning The structured representation of 3D scene graphs has been predominantly employed for task planning by using their hierarchical structure [ 11 , 19 , 13 ] , and even incorporating natural language processing [ 12 , 20 , 21 ] by using LLM ( LLM )s to match keywords in human-intuitive commands to nodes in the scene graph. Other works have utilized scene graphs to reason [ 22 ] , infer [ 21 , 23 ] , or learn [ 24 ] navigation actions and policies. However, their application to low-level path planning in geometric spaces remains highly underexplored. Recently, HOV-SG [ 25 ] processes natural language queries across three levels of abstraction (floor, room, and object) to exploit the hierarchical structure of a scene graph and progressively narrow down the solution space. However, this is only used to identify the goal node for navigation; the path planning still relies

Read the complete text