Research Analyzer
← Back ICRA 2026

Terra: Hierarchical Terrain-Aware 3D Scene Graph for Task-Agnostic Outdoor Mapping

Chad Samuelson, Abigail Austin, Seth Knoop, Blake Romrell, Gabriel Slade, T.W. McLain, Joshua Mangelson

PDF

AI summary

Key figure (auto-extracted from paper)
Terra enables memory-efficient, task-agnostic outdoor mapping by combining LiDAR SLAM with terrain-aware hierarchical scene graphs, outperforming state-of-the-art methods in region classification while remaining lightweight.
outdoor mapping 3D scene graphs LiDAR SLAM terrain-aware navigation metric-semantic mapping autonomous robotics

Problem

Existing 3D scene graph methods are largely confined to indoor environments, rely on computationally intensive mesh reconstruction, or are task-dependent, leaving a gap for scalable, terrain-aware, and open-set semantic mapping in large outdoor settings.

Approach

Terra integrates LiDAR SLAM with vision-language models to build a sparse metric-semantic map, then structures it into a hierarchical scene graph with terrain-aware place nodes and multi-level region clusters for flexible, task-driven querying.

Key results

  • Memory-efficient, task-agnostic open-set metric-semantic mapping for large outdoor environments
  • Terrain-aware place nodes and hierarchical region layers that improve region classification over SOTA indoor 3DSG methods
  • On-par performance with state-of-the-art camera-based 3DSG methods in object retrieval tasks
  • Successful demonstration in diverse robotic tasks across simulation and real-world wheelchair platforms

Why it matters

It provides outdoor autonomous robots with a lightweight, scalable semantic map that improves terrain-aware navigation and task execution without the computational burden of indoor mesh-based methods.

Abstract

Outdoor intelligent autonomous robotic operation relies on a sufficiently expressive map of the environment. Classical geometric mapping methods retain essential structural environment information, but lack a semantic understanding and organization to allow high-level robotic reasoning. 3D scene graphs (3DSGs) address this limitation by integrating geomet- ric, topological, and semantic relationships into a multi-level graph-based map. Outdoor autonomous operations commonly rely on terrain information either due to task-dependence or the traversability of the robotic platform. We propose a novel approach that combines indoor 3DSG techniques with standard outdoor geometric mapping and terrain-aware reasoning, pro- ducing terrain-aware place nodes and hierarchically organized regions for outdoor environments. Our method generates a task-agnostic metric-semantic sparse map and constructs a 3DSG from this map for downstream planning tasks, all while remaining lightweight for autonomous robotic operation. Our thorough evaluation demonstrates our 3DSG method performs on par with state-of-the-art camera-based 3DSG methods in object retrieval and surpasses them in region classification while remaining memory efficient. We demonstrate its effectiveness in diverse robotic tasks of object retrieval and region monitoring in both simulation and real-world environments. Our Github: https://github.com/BYU-FROST-Lab/Terra.

Index terms

Semantic Scene Understanding Object Detection Segmentation and Categorization Mapping

Related papers