Dungeon and Visibility World Generator

Edwin Mériaux; Antonio Loría; Gregory Dudek

cs.ROOct 14, 2026

Dungeon and Visibility World Generator

Authors: Edwin Mériaux, Antonio Loría, Gregory Dudek

Organizations: McGill University, Montréal QC H3A 0G4, Canada · Université Paris-Saclay, 91190 Gif-sur-Yvette, France · CNRS, Laboratoire des Signaux et Systèmes (L2S), 91190 Gif-sur-Yvette, France

Abstract

Robotic exploration, coverage, and multi-agent deployment algorithms require reproducible environments with controllable geometry and visibility. This paper presents the Dungeon and Visibility World Generator, an open-source framework for constructing two- and three-dimensional environments and evaluating geometric line-of-sight. The framework supports orthogonal, polygonal or polyhedral, and curvilinear geometries, with six construction families covering scattered rooms, clusters, winding passages, caverns, branches, and loops. Room dimensions, passage structures, obstacles, and spatial placement can be controlled through configurable parameters and probability distributions. Each generated environment provides continuous free-space geometry and a corresponding discrete occupancy representation. A Rust-based visibility engine supports line-of-sight queries in both representations, enabling the evaluation of observable regions from arbitrary agent positions. The framework records configurations, generated geometries, occupancy grids, and visibility results to support reproducible experiments. We demonstrate the generator across all six geometry classes, examine how geometry and spatial resolution affect visibility, and evaluate generation time and reproducibility. The resulting framework provides a configurable foundation for evaluating robotic exploration, coverage, and multi-agent deployment algorithms in diverse environments.

Figures & tables

Figure 2: The six geometry classes, illustrated by saved worlds. Top: orthogonal, polygonal and curvilinear 2D worlds, each in a $50\times 50$ -unit domain. Bottom: orthogonal, polyhedral and curvilinear 3D worlds in $50\times 50\times 30$ -unit domains. White regions are planar free space and dark islands are obstacles. Translucent blue surfaces show 3D free-space boundaries, with dark interior obstacles. The 3D meshes are display approximations; the saved solid defines membership and visibility. The panels use different layouts and illustrate classes rather than a comparison with layout held fixed.

Figure 3: One saved construction layout for each family, before class-specific geometric realization. White outlined shapes are room templates, orange points are their saved centers, teal lines are passage centerlines, and pale teal bands show the recorded passage widths. These are construction elements before union; overlapping rooms or paths can merge in the final world. Inserted obstacles are omitted. All panels use a

50\times 50

-unit planar domain. The 3D realization assigns elevations and adds inter-elevation connections to the planar layout.

Figure 4: Three stages for the same saved 2D world. Left: room templates, centers and passage paths. Middle: the constructed polygon with finite-width passages and blocked holes. Right: its conservative grid has one connected free component. Checking accepts or rejects a candidate; it does not alter its geometry.

Figure 5: A 2D illustration of whole-cell safety. Left: the entire unit square lies in free space;

H(c)=1>\rho_{\mathrm{2D}}

. Right: its center is free, but part of the square crosses the wall. The enclosing circle illustrates the clearance principle used with a sphere in 3D; the actual 2D implementation tests whole-square containment directly.

Figure 6: The same window of a saved 2D world. Left: continuous sample points. Right: one target per free cell, with spacing 0.5. Orange stars show one continuous observer and its assigned free grid center.

Class	Family	Rooms/passages	Free entries	Median (s)	Range (s)
2D Orthogonal	Scattered rooms	10/12	1,775	0.027	0.026–0.033
2D Polygonal	Scattered rooms	10/12	1,127	0.025	0.025–0.032
2D Curvilinear	Winding passages	8/5	1,503	0.029	0.028–0.030
3D Orthogonal	Room clusters	9/10	18,512	1.909	1.888–1.950
3D Polyhedral	Scattered rooms	10/12	8,179	2.853	2.781–3.308
3D Curvilinear	Branches	9/8	8,409	3.162	3.144–3.344

Table 1: Complete generation-call times for one fixed world per class. Medians and minimum–maximum ranges use five timed repeats after one excluded warm-up per configuration. Rooms/passages count the realized construction elements before their union. All configurations use seed 42. Full timings and configurations are archived with the study.

Compared artifact	Timed comparisons	Fresh-process replay
geometry.geojson or solid.json	24/24 match	6/6 match
free.npy	24/24 match	6/6 match
layout.json	24/24 match	6/6 match
Requested and resolved configurations	24/24 match	6/6 match
connectivity_certificate.json	24/24 match	6/6 match
primitives.geojson (2D only)	12/12 match	3/3 match

Table 2: Reproduction checks against each world’s first timed run. There are six baselines, 24 additional timed comparisons and six fresh-process comparisons. No geometric or coordinate tolerance is used to declare artifact equality.

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Panel of Figure 27 — Figure 27: Successful-method panels from Figures 16–18 of Meriaux et al. [ 2 ] . Original vector figure files are retained; only the unsuccessful-method panels are cropped. The plotted values and shaded regions are not recomputed or fitted here. Provenance records the versioned source and file hashes.

Procedural Content Generation

Procedural content generation creates game levels, 3D worlds, assets, stories, and simulation environments through algorithms and learned models.

See all 67 papers

D3D-GEN: Robot-Aware Domain-Grounded Interactive 3D World Generation for Social Robotics
Aug 12, 2026Anh Duc Do, Volodymyr Shcherbyna, Tai Duc Nguyen +5Robotics Simulation Functional 3D Scene Generation
Singapore Management University · Max Planck Institute · Technical University Braunschweig
MANSION: Multi-floor lANguage-to-3D Scene generatIOn for loNg-horizon tasks
Mar 12, 2026Lirong Che, Shuo Wen, Shan Huang +53D Scene Generation Indoor Scene Generation
Tsinghua University · AgiBot · McGill University, MILA - Quebec AI Institute
WorldRover: A Scalable Synthetic Video Data Engine for World Exploration with Rich Annotations
Aug 16, 2026Xiaojie Xu, Zhengyuan Lin, Runyi Li +3Synthetic Data Generation Video World Models
Alaya Lab · The University of Tokyo

Aug 12, 2026J/K move · Enter open · S save

D3D-GEN: Robot-Aware Domain-Grounded Interactive 3D World Generation for Social Robotics

Anh Duc Do, Volodymyr Shcherbyna, Tai Duc Nguyen, Spaarsh Thakkar+4

Singapore Management University · Max Planck Institute · Technical University Braunschweig

Robotics Simulation Functional 3D Scene Generation 3D Scene Layout Generation Indoor Scene Generation Social Robotics

Open paper AskPDF arXiv

Training and validation of Embodied AI for social navigation critically depends on realistic simulation environments, yet many current approaches fail to find a balance between realism and simulability. We propose D3D-GEN, a novel world generation system that combines a domain agent with a retrieval-augmented generation (RAG) pipeline grounded in that domain. Our system enables users to rapidly generate domain-grounded, fully interactive 3D worlds by automating both the collection of domain knowledge and the synthesis of realistic floorplans and object placements, without dependence on any fixed 3D model database. Given a domain description prompt, the research agent collects publicly accessible domain-specific data and constructs a persistent domain database. Using this database, our RAG pipeline generates plausible floorplans and object placements by dynamically querying a user-provided semantic database, which can be easily extended or modified. The output is a fully interactive 3D world loadable by the popular simulators Isaac Sim and Gazebo. With our approach, we have built databases for several common domains (indoor residential, hospital, office) and generated dozens of distinct, plausible simulation environments for each domain. We present D3D-GEN with a local web frontend that facilitates rapid, interactive world generation for robot simulation.

Abstract

Procedural Content Generation

Explore similar work

D3D-GEN: Robot-Aware Domain-Grounded Interactive 3D World Generation for Social Robotics

MANSION: Multi-floor lANguage-to-3D Scene generatIOn for loNg-horizon tasks

WorldRover: A Scalable Synthetic Video Data Engine for World Exploration with Rich Annotations