pith. sign in

arxiv: 2605.20917 · v1 · pith:PUWJKZC3new · submitted 2026-05-20 · 💻 cs.RO

SubTGraph: Large-Scale Subterranean Environment Synthesis with Controllable Topological Variability for Robotic Autonomy Validation

F. Labra Caso , A. Saradagi , S. Fredriksson , S. Nordstr\"om , A. Koval , G. Nikolakopoulos This is my paper

Pith reviewed 2026-05-21 04:45 UTC · model grok-4.3

classification 💻 cs.RO
keywords subterranean environmentsrobotic autonomy validationprocedural world generationtopological variabilitysimulation benchmarkingpath planningSLAMenvironment synthesis
0
0 comments X

The pith

SubTGraph creates many distinct subterranean environments from user-specified constraints using cost matrices and Dijkstra to support statistical testing of robot autonomy.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

SubTGraph is a framework that turns user rules on topology, size, and appearance into procedural underground worlds at large scale. It assembles topometric tiles from the DARPA generator by running Dijkstra on a cost matrix built from those rules. This addresses the common practice of validating robot systems in only a few simulation spaces, which limits how much one can trust the results. The method produces hundreds of varied settings such as mines, caves, and lava tubes. The authors demonstrate its use for testing segmentation, path planning, and SLAM, and release the code plus 150 example worlds.

Core claim

SubTGraph builds a cost matrix from user-specified structural constraints to guide the classical Dijkstra algorithm to procedurally generate SubT worlds utilizing topometric tiles from the DARPA World Generator. This produces multi-level environments with controllable variability in topology and other features, enabling the creation of distinct settings like operational mines, natural caves, and lava tubes for rigorous autonomy validation.

What carries the argument

A cost matrix derived from user constraints that directs Dijkstra's shortest-path algorithm to select and connect topometric tiles into coherent subterranean graphs.

If this is right

  • Robotic semantic segmentation can be evaluated against reliable topometric ground truth across many different environments.
  • Multi-agent path planning algorithms can be run on hundreds of topologies to reveal consistent patterns or failure modes.
  • LIO-based SLAM systems can be tested in difficult underground sections to locate specific failure cases.
  • The open-sourced database of 150 worlds provides a ready benchmark set for statistical evaluation of autonomy stacks.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • This generation method might be adapted to create variable environments in other domains, such as urban disaster sites or extraterrestrial surfaces, given appropriate base tiles.
  • Large sets of generated worlds could serve as training data for learning-based robot controllers that need to handle topological diversity.
  • Community extensions to the open-source code could add new constraint types or tile libraries to increase the range of possible worlds.

Load-bearing premise

That the topometric tiles and the constraint-guided Dijkstra procedure can generate worlds that are realistic and diverse enough to replace or supplement real-world and hand-crafted test environments for autonomy validation.

What would settle it

A direct comparison in which autonomy performance metrics collected over the 150 generated worlds fail to show statistically significant differences from results obtained in a small number of manually designed spaces.

Figures

Figures reproduced from arXiv: 2605.20917 by A. Koval, A. Saradagi, F. Labra Caso, G. Nikolakopoulos, S. Fredriksson, S. Nordstr\"om.

Figure 1
Figure 1. Figure 1: SubTGraph Conceptual Diagram. User requirements specify control over the topology, dimensionality and [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The algorithm creates 2-dimensional topologies for each specified level by utilizing the route descriptors [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: This figure displays the recursive generation process of the mesh object. Once the constraint distribution [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Single-level Occupancy Matrix. Two intersec [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: This figure presents the three subterranean environment types generated by the SubTGraph software: (1) [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The object spawn and interior textures are controlled by the user specifications. These can be extended to [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Similarity Heatmap of environments on linear, parabolic and sine circuits. [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: (a) Symmetry of environments with horizontal, vertical and rotational mirroring. (b) Asset appearance [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: (a) Mesh generation time on every environment type for system configurations in Table 2. (b) Virtual RAM [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Environment size distribution on dataset bench [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Topometric map estimated through GRID￾FAST over an operational mine. The map is transformed into structural semantic components such as intersections and pathways [PITH_FULL_IMAGE:figures/full_fig_p010_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Visualization of the Gazebo physics simulator [PITH_FULL_IMAGE:figures/full_fig_p010_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: A∗ +T multi-agent path planning over an unstruc￾tured natural cave. The method accounts for obstacles and agent path conflicts through temporal estimation. 5.2 Multi-robot Path Planning Another application of the SubTGraph creation tool is multi-robot path planning. This study employs the multi￾agent path planner A∗ +T [21] to estimate a global path planning utilizing established maps of natural cave envi… view at source ↗
Figure 14
Figure 14. Figure 14: Evaluation of (a) FastLIO and (b) DLIO during an aerial exploration mission for the case of vertical shaft [PITH_FULL_IMAGE:figures/full_fig_p011_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Evaluation of (a) FastLIO and (b) DLIO during [PITH_FULL_IMAGE:figures/full_fig_p012_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: GRID-FAST validation on dataset subsection. Black color indicates straight areas, yellow sections [PITH_FULL_IMAGE:figures/full_fig_p014_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: A∗ +T validation on dataset subsection. Three agents are spawned with an associated objective position on different sections of the natural cave. The optimization process searches for the fastest route while evading other agents’ paths. Results on (d),(e),(j) display overlapping paths, showing A∗ +T’s capability to utilize the same space for multiple agents at different times. 15 [PITH_FULL_IMAGE:figures… view at source ↗
Figure 18
Figure 18. Figure 18: LIO algorithm validation. Benchmarking on different sections of the generated underground worlds, two [PITH_FULL_IMAGE:figures/full_fig_p016_18.png] view at source ↗
read the original abstract

Subterranean (SubT) environments have been a frontier for autonomous robotics, driven by the push for automation of mining operations and the interest in planetary exploration (Martian Lava Tubes). Due to the challenges involved in accessing real SubT environments, rigorous hardening of autonomy stacks in realistic simulation environments is critical. This article fills a well-known gap, which relates to the unavailability of a large-scale simulation-based benchmarking infrastructure for rigorous statistical evaluation of robotic autonomy, due to which it is common for SubT research articles to present validation results in a few environments at best. This article presents SubTGraph, a novel framework for rapid synthesis of multi-level SubT environments with high variability, incorporating user specifications related to topology, dimensionality, textures, etc., to generate distinct environments such as operational mines, natural caves and lava tubes. SubTGraph builds a cost matrix from user-specified structural constraints to guide the classical Dijkstra algorithm to procedurally generate SubT worlds utilizing topometric tiles from the DARPA World Generator. Three robotics case-studies are investigated to demonstrate the utility of SubTGraph for rigorous validation of different layers in the robotic autonomy stack. Structural semantic segmentation is validated against topometric ground truths, multi-agent path planning is widely tested for identification of patterns and trends in the algorithm behavior and LIO SLAM is stress-tested in challenging subterranean sections to identify failure cases. The SubTGraph world creation codebase is open-sourced (https://github.com/LTU-RAI/SubTGraph.git) along with a database consisting of 150 highly variable underground worlds.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript presents SubTGraph, a procedural framework for rapid synthesis of multi-level subterranean environments with controllable topological variability. It constructs a cost matrix from user-specified structural constraints and applies the Dijkstra algorithm to assemble topometric tiles from the DARPA World Generator, producing 150 distinct worlds (operational mines, natural caves, lava tubes). Utility is shown via three case studies validating structural semantic segmentation against topometric ground truth, identifying behavioral patterns in multi-agent path planning, and stress-testing LIO-SLAM for failure cases, with the generation codebase and database released openly.

Significance. If the tile-based environments are shown to produce representative failure modes for segmentation, planning, and SLAM, the work would meaningfully address the scarcity of large-scale, statistically rigorous benchmarking infrastructure in SubT robotics. The open-sourced tool and 150-world database constitute a concrete, reusable contribution that could support community-wide validation efforts.

major comments (2)
  1. [Abstract] Abstract: the three case studies are described without any quantitative metrics, error bars, statistical tests, or details on environment sampling and failure-mode identification; this directly weakens the central claim that SubTGraph enables rigorous statistical evaluation of autonomy stacks.
  2. [World-generation pipeline] World-generation pipeline (described after the cost-matrix construction): the claim that the resulting layouts support representative autonomy validation rests on the untested premise that topological variability from DARPA tiles plus user textures suffices; no comparison to real SubT statistics (fractal roughness, cross-section variability, stochastic occlusions) is provided, which is load-bearing for the realism needed in the case studies.
minor comments (1)
  1. [Figures] Figure captions and legends could more explicitly indicate which user constraints were active for each example world.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment point-by-point below, indicating revisions where appropriate to strengthen the presentation of SubTGraph's contributions for statistical validation of robotic autonomy.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the three case studies are described without any quantitative metrics, error bars, statistical tests, or details on environment sampling and failure-mode identification; this directly weakens the central claim that SubTGraph enables rigorous statistical evaluation of autonomy stacks.

    Authors: We acknowledge that the abstract provides only a high-level description of the case studies. The full manuscript reports quantitative results for each: segmentation accuracy metrics against topometric ground truth across sampled environments, statistical patterns and trends (including variability measures) in multi-agent path planning over the 150 worlds, and specific failure rates and conditions identified for LIO-SLAM in challenging sections. To better support the central claim within the abstract's constraints, we will revise it to include key quantitative highlights, sampling details, and references to the statistical nature of the evaluations. revision: yes

  2. Referee: [World-generation pipeline] World-generation pipeline (described after the cost-matrix construction): the claim that the resulting layouts support representative autonomy validation rests on the untested premise that topological variability from DARPA tiles plus user textures suffices; no comparison to real SubT statistics (fractal roughness, cross-section variability, stochastic occlusions) is provided, which is load-bearing for the realism needed in the case studies.

    Authors: SubTGraph assembles environments from topometric tiles provided by the DARPA World Generator, which were developed specifically to represent real subterranean settings from the DARPA Subterranean Challenge. The cost-matrix and Dijkstra procedure then impose controllable topological variability on top of these tiles according to user constraints. While the manuscript does not include a direct quantitative comparison of generated layouts against real-world SubT statistics such as fractal roughness or cross-section variability, the three case studies demonstrate that the resulting environments produce relevant and representative autonomy challenges (e.g., segmentation errors in complex topologies and SLAM drift in occluded areas). We therefore maintain that the framework supports statistical validation; a dedicated statistical realism analysis would be a valuable extension but lies beyond the scope of the current work focused on synthesis and application. revision: no

Circularity Check

0 steps flagged

No circularity: procedural synthesis pipeline is self-contained

full rationale

The paper describes a procedural world-generation method that constructs a cost matrix from user-specified structural constraints and then applies the standard Dijkstra algorithm to select and assemble topometric tiles sourced from the external DARPA World Generator. No equations, predictions, or derivations are presented that reduce to fitted parameters or self-referential definitions. The central claim (generation of 150 variable SubT environments for autonomy validation) rests on the composition of independent external components and a classical graph algorithm rather than any internal loop or self-citation chain. Validation case studies are presented as demonstrations, not as inputs that define the generation process. This is a standard non-circular engineering pipeline.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The framework rests on the availability and representativeness of the external DARPA tile set and on the assumption that constraint-to-cost-matrix translation produces useful variability; no new physical entities or fitted constants are introduced in the abstract.

axioms (1)
  • domain assumption Topometric tiles from the DARPA World Generator can be combined to represent diverse real-world subterranean structures.
    Invoked when the generation pipeline is described as utilizing these tiles to create operational mines, caves, and lava tubes.

pith-pipeline@v0.9.0 · 5847 in / 1344 out tokens · 37019 ms · 2026-05-21T04:45:37.402427+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

24 extracted references · 24 canonical work pages

  1. [1]

    A review of physics simula- tors for robotic applications.IEEE Access, 9:51416– 51431, 2021

    Jack Collins, Shelvin Chand, Anthony Vanderkop, and David Howard. A review of physics simula- tors for robotic applications.IEEE Access, 9:51416– 51431, 2021

    work page 2021

  2. [2]

    A subterranean virtual cave world for gazebo based on the darpa subt challenge.arXiv preprint arXiv:2004.08452, 2020

    Anton Koval, Christoforos Kanellakis, Emil Vid- mark, Jakub Haluska, and George Nikolakopou- los. A subterranean virtual cave world for gazebo based on the darpa subt challenge.arXiv preprint arXiv:2004.08452, 2020

    work page arXiv 2004

  3. [3]

    Uncertainty and variability in point cloud surface data

    Mark Pauly, Niloy J Mitra, and Leonidas J Guibas. Uncertainty and variability in point cloud surface data. InEurographics symposium on point-based graphics, 2004

    work page 2004

  4. [4]

    Subjective and objective quality evaluation of 3d point cloud denoising algorithms

    Alireza Javaheri, Catarina Brites, Fernando Pereira, and João Ascenso. Subjective and objective quality evaluation of 3d point cloud denoising algorithms. In2017 IEEE International Conference on Multime- dia & Expo Workshops (ICMEW), pages 1–6. IEEE, 2017

    work page 2017

  5. [5]

    Surface reconstruction from point clouds: A survey and a benchmark.IEEE transac- tions on pattern analysis and machine intelligence, 2024

    Zhangjin Huang, Yuxin Wen, Zihao Wang, Jinjuan Ren, and Kui Jia. Surface reconstruction from point clouds: A survey and a benchmark.IEEE transac- tions on pattern analysis and machine intelligence, 2024. 12 ARXIVPREPRINT05-2026

    work page 2024

  6. [6]

    Screened pois- son surface reconstruction.ACM Transactions on Graphics (ToG), 32(3):1–13, 2013

    Michael Kazhdan and Hugues Hoppe. Screened pois- son surface reconstruction.ACM Transactions on Graphics (ToG), 32(3):1–13, 2013

    work page 2013

  7. [7]

    Deep implicit mov- ing least-squares functions for 3d reconstruction

    Shi-Lin Liu, Hao-Xiang Guo, Hao Pan, Peng-Shuai Wang, Xin Tong, and Yang Liu. Deep implicit mov- ing least-squares functions for 3d reconstruction. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 1788– 1797, 2021

    work page 2021

  8. [8]

    Robots conquer the underground: What darpa’s subterranean challenge means for the fu- ture of autonomous robots.IEEE Spectrum, 59(5):30– 37, 2022

    Evan Ackerman. Robots conquer the underground: What darpa’s subterranean challenge means for the fu- ture of autonomous robots.IEEE Spectrum, 59(5):30– 37, 2022

    work page 2022

  9. [9]

    Integrating grid- based and topological maps for mobile robot naviga- tion

    Sebastian Thrun and Arno Bücken. Integrating grid- based and topological maps for mobile robot naviga- tion. InProceedings of the national conference on artificial intelligence, pages 944–951. Citeseer, 1996

    work page 1996

  10. [10]

    Communications in a cave environment

    William Walsh and Jay Gao. Communications in a cave environment. In2018 IEEE Aerospace Confer- ence, pages 1–8. IEEE, 2018

    work page 2018

  11. [11]

    Topological exploration of subter- ranean environments.Journal of Field Robotics, 23(6-7):395–415, 2006

    David Silver, Dave Ferguson, Aaron Morris, and Scott Thayer. Topological exploration of subter- ranean environments.Journal of Field Robotics, 23(6-7):395–415, 2006

    work page 2006

  12. [12]

    Au- tonomous system for navigation and surveying in underground mines.Journal of Field Robotics, 24(10):829–847, 2007

    Joseph Nsasi Bakambu and Vladimir Polotski. Au- tonomous system for navigation and surveying in underground mines.Journal of Field Robotics, 24(10):829–847, 2007

    work page 2007

  13. [13]

    What localizes beneath: A metric multisensor localization and mapping system for autonomous underground mining vehicles.Journal of Field Robotics, 38(1):5– 27, 2021

    Adam Jacobson, Fan Zeng, David Smith, Nigel Boswell, Thierry Peynot, and Michael Milford. What localizes beneath: A metric multisensor localization and mapping system for autonomous underground mining vehicles.Journal of Field Robotics, 38(1):5– 27, 2021

    work page 2021

  14. [14]

    Design of an autonomous robot for mapping, navigation, and manipulation in underground mines

    Robert Lösch, Steve Grehl, Marc Donner, Claudia Buhl, and Bernhard Jung. Design of an autonomous robot for mapping, navigation, and manipulation in underground mines. In2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 1407–1412. IEEE, 2018

    work page 2018

  15. [15]

    Into the robotic depths: Analysis and insights from the darpa subterranean challenge.Annual Re- view of Control, Robotics, and Autonomous Systems, 6(1):477–502, 2023

    Timothy H Chung, Viktor Orekhov, and Angela Maio. Into the robotic depths: Analysis and insights from the darpa subterranean challenge.Annual Re- view of Control, Robotics, and Autonomous Systems, 6(1):477–502, 2023

    work page 2023

  16. [16]

    Autonomous multi-robot exploration with ground vehicles in darpa subterranean challenge finals.Field Robotics, 3:266– 300, 2023

    Jan Bayer, Petr Cížek, and Jan Faigl. Autonomous multi-robot exploration with ground vehicles in darpa subterranean challenge finals.Field Robotics, 3:266– 300, 2023

    work page 2023

  17. [17]

    On the use of simulation in robotics: Opportuni- ties, challenges, and suggestions for moving forward

    HeeSun Choi, Cindy Crump, Christian Duriez, Asher Elmquist, Gregory Hager, David Han, Frank Hearl, Jessica Hodgins, Abhinandan Jain, Frederick Leve, et al. On the use of simulation in robotics: Opportuni- ties, challenges, and suggestions for moving forward. Proceedings of the National Academy of Sciences, 118(1):e1907856118, 2021

    work page 2021

  18. [18]

    Grid-fast: A grid-based intersection de- tection for fast semantic topometric mapping.Jour- nal of Intelligent & Robotic Systems, 110(4):154, 2024

    Scott Fredriksson, Akshit Saradagi, and George Niko- lakopoulos. Grid-fast: A grid-based intersection de- tection for fast semantic topometric mapping.Jour- nal of Intelligent & Robotic Systems, 110(4):154, 2024

    work page 2024

  19. [19]

    Robotic Exploration through Semantic Topometric Mapping

    Scott Fredriksson, Akshit Saradagi, and George Niko- lakopoulos. Robotic Exploration through Semantic Topometric Mapping. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 9404–9410, May 2024

    work page 2024

  20. [20]

    Multi-Agent Path Find- ing Using Conflict-Based Search and Structural- Semantic Topometric Maps

    Scott Fredriksson, Yifan Bai, Akshit Saradagi, and George Nikolakopoulos. Multi-Agent Path Find- ing Using Conflict-Based Search and Structural- Semantic Topometric Maps. pages 4229–4235, May 2025

    work page 2025

  21. [21]

    A time-dependent risk- aware distributed multi-agent path finder based on a

    Samuel Nordström, Yifan Bai, B Lindqvist, and George Nikolakopoulos. A time-dependent risk- aware distributed multi-agent path finder based on a. arXiv preprint arXiv:2504.19593, 2025

    work page arXiv 2025

  22. [22]

    D+*: A risk aware platform agnostic heterogeneous path planner

    Samuel Karlsson, Anton Koval, Christoforos Kanel- lakis, and George Nikolakopoulos. D+*: A risk aware platform agnostic heterogeneous path planner. Expert Systems with Applications, 215:119408, 2023

    work page 2023

  23. [23]

    Fast-lio2: Fast direct lidar-inertial odom- etry.IEEE Transactions on Robotics, 38(4):2053– 2073, 2022

    Wei Xu, Yixi Cai, Dongjiao He, Jiarong Lin, and Fu Zhang. Fast-lio2: Fast direct lidar-inertial odom- etry.IEEE Transactions on Robotics, 38(4):2053– 2073, 2022

    work page 2053

  24. [24]

    Direct lidar-inertial odometry: Lightweight lio with continuous-time motion correction.arXiv preprint arXiv:2203.03749, 2022

    Kenny Chen, Ryan Nemiroff, and Brett T Lopez. Direct lidar-inertial odometry: Lightweight lio with continuous-time motion correction.arXiv preprint arXiv:2203.03749, 2022. 13 ARXIVPREPRINT05-2026 A Method Validation on SubTGraph Benchmark Dataset A.1 GRID-FAST Topometric Validation Figure 16: GRID-FAST validation on dataset subsection. Black color indicat...

    work page arXiv 2022