TreeLoc++: Robust 6-DoF LiDAR Localization in Forests with a Compact Digital Forest Inventory
Minwoo Jung 1 {}^{\textbf{1}}, Dongjae Lee 1 {}^{\textbf{1}}, Nived Chebrolu 2 {}^{\textbf{2}}, Haedam Oh 2 {}^{\textbf{2}}, Maurice Fallon 2 {}^{\textbf{2}} and Ayoung Kim 1 {}^{\textbf{1}} · 2026
Read TreeLoc++: Robust 6-DoF LiDAR Localization in Forests with a Compact Digital Forest Inventory by Minwoo Jung 1 {}^{\textbf{1}}, Dongjae Lee 1 {}^{\textbf{1}}, Nived Chebrolu 2 {}^{\textbf{2}}, Haedam Oh 2 {}^{\textbf{2}}, Maurice Fallon 2 {}^{\textbf{2}} and Ayoung Kim 1 {}^{\textbf{1}} (2026) 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-2603-03695.txt.
Introduction
Minwoo Jung and five colleagues locate robots in forests using a phone-book of trees instead of a mountain of laser data. Forest inventories already catalog stems by compact geometric traits, yet localization methods ignore them and haul around gigabyte point clouds. Their 2026 system, TreeLoc++, works directly on the inventory: a list of tree positions and trunk diameters becomes the whole map.
Coarse retrieval is hardened with a histogram of pairwise tree distances, which encodes the local layout and cuts false matches in look-alike forest. Diameter filtering and a spin-consistency check prune candidates further, and a constrained optimization settles the full six-degree pose. Across twenty-seven sequences in four countries the system localized to centimeter accuracy. Fifteen sessions covering about eight kilometers of trails fit in a map smaller than a single photograph, and data localized two years after the inventory was built.
Catalog record
| CREATOR | Minwoo Jung 1 {}^{\textbf{1}}, Dongjae Lee 1 {}^{\textbf{1}}, Nived Chebrolu 2 {}^{\textbf{2}}, Haedam Oh 2 {}^{\textbf{2}}, Maurice Fallon 2 {}^{\textbf{2}} and Ayoung Kim 1 {}^{\textbf{1}} |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 114455 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.03695
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
TreeLoc++: Robust 6-DoF LiDAR Localization in Forests with a Compact Digital Forest Inventory Minwoo Jung 1 {}^{\textbf{1}}, Dongjae Lee 1 {}^{\textbf{1}}, Nived Chebrolu 2 {}^{\textbf{2}}, Haedam Oh 2 {}^{\textbf{2}}, Maurice Fallon 2 {}^{\textbf{2}} and Ayoung Kim 1 {}^{\textbf{1}} Abstract Reliable localization is essential for sustainable forest management, as it allows robots or sensor systems to revisit and monitor the status of individual trees over long periods. In modern forestry, this management is structured around Digital Forest Inventories (DFIs), which encode stems using compact geometric attributes rather than raw data. Despite their central role, DFIs have been overlooked in localization research, and most methods still rely on dense gigabyte-sized point clouds that are costly to store and maintain. To improve upon this, we propose TreeLoc++, a global localization framework that operates directly on DFIs as a discriminative representation, eliminating the need to use the raw point clouds. TreeLoc++ substantially strengthens the existing TreeLoc framework by reducing false matches in structurally ambiguous forests and improving the reliability of full 6-DoF pose estimation. Building upon TreeLoc, it augments coarse retrieval with a pairwise distance histogram that encodes local tree-layout context, subsequently refining candidates via DBH-based filtering and yaw-consistent inlier selection to further reduce mismatches. Furthermore, a constrained optimization leveraging tree geometry jointly estimates roll, pitch, and height, enhancing pose stability and enabling accurate localization without reliance on dense 3D point cloud data. Evaluations on 27 sequences recorded in forests across three datasets and four countries show that TreeLoc++ achieves precise localization with centimeter-level accuracy. We further demonstrate robustness to long-term change by localizing data recorded in 2025 against inventories built from 2023 data, spanning a two-year interval. The system represents 15 sessions spanning 7.98 km of trajectories using only 250 KB of map data and outperforms both hand-crafted and learning-based baselines that rely on dense point cloud maps. This demonstrates the scalability of TreeLoc++ for long-term deployment. Code and datasets will be released to support future research. \IEEEoverridecommandlockouts \overrideIEEEmargins {IEEEkeywords} Environmental Monitoring, Global Localization, Place Recognition, Pose Estimation † † corresponding: Corresponding author: Ayoung Kim (email: [email protected]). † † note: This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. RS-2024-00461409); the Horizon Europe project DigiForest (101070405); and EPSRC Project Mobile Robotic Inspector (EP/Z531212/1). 1 Introduction \IEEEPARstart R eliable and accurate localization is essential in forestry and environmental monitoring, where systems operate in complex under-canopy environments [ 76 , 47 , 46 ] . Localization enables key long-term tasks such as repeated site access, consistent map maintenance under dynamic canopy conditions, and digital inventory updates for sustainable forest management [ 45 , 33 , 52 ] . For practical deployment, localization systems must be robust to challenges specific to forests such as structural repetition and seasonal variation, provide centimeter-level accuracy, and rely on input representations that minimize storage and computational overhead in long-term, resource-constrained missions [ 6 , 37 ] . Figure 1 : (Left) Images captured from the same location in the Evo forest two years apart, showing vegetation growth (yellow) and denser foliage (cyan), which amplifies perceptual aliasing. (Right) TreeLoc++ achieves accurate multi-session alignment using only lightweight DFIs by leveraging tree attributes, avoiding raw point clouds that can vary across sessions due to different LiDAR sensors (black). While satellite-based localization systems such as GNSS are widely used in outdoor environments, they struggle in forests due to signal degradation under dense canopies [ 65 ] . Even high-precision RTK services often fail to maintain consistent performance [ 30 ] , resulting in meter-level errors, which are insufficient for tasks requiring tree-level correspondence such as growth estimation [ 9 ] . These limitations underscore the need for onboard, geometry-based alternatives that do not rely on external positioning infrastructure; LiDAR is a promising solution due to its ability to capture precise 3D geometric information [ 7 , 71 ] . The standard LiDAR-based global localization pipeline [ 76 ] typically involves place recognition for candidate retrieval [ 25 , 24 , 26 ] followed by pose estimation through descriptor matching [ 78 , 77 , 17 ] or scan registration [ 75 , 36 , 35 ] . While effective in structured environments, these methods incur high memory usage due to dense point cloud storage while localization performance degrades in forests, where geometrically similar scenes and seasonal variation cause perceptual aliasing [ 48 ] . This leads to more frequent retrieval errors and unstable pose estimates. To mitigate reliance on dense point clouds and reduced discriminative power in forests, recent research has proposed compact, high-level representations as alternatives to raw point clouds [ 31 , 10 ] . Segmentation-based methods reduce storage by clustering geometrically or semantically similar points, but their discriminative power remains limited in forests due to low semantic diversity, where only a few semantic categories recur [ 66 ] . Other methods such as ESM [ 62 ] and NSM [ 61 ] extract descriptors from individual tree instances, but still rely on point clouds and overlook inter-tree relationships, making them prone to perceptual aliasing in environments where tree geometries are indistinguishable. These limitations motivate the use of more compact and interpretable representations that are better suited to the structural characteristics of forest environments. A more compact alternative would be to use a DFI ( DFI ) directly for localization. A DFI represents each tree by attributes such as stem axis and DBH ( DBH ) without requiring any raw point clouds. TreeLoc [ 22 ] introduced a DFI-based framework that leverages inter-tree spatial relationships for matching. While efficient, it remains sensitive to structural aliasing, where locally similar tree layouts repeat across different parts of the forest, leading to incorrect matches and localization failures. To overcome these limitations, we propose TreeLoc++, a forest-specific localization system that builds on TreeLoc to address structural ambiguity and false correspondence matches, improving candidate retrieval and enabling reliable 6-DoF pose estimation. TreeLoc++ enhances the original coarse-to-fine framework with components that improve robustness and efficiency. It augments the TDH ( TDH ) introduced in TreeLoc, a histogram-based descriptor encoding individual tree attributes, with a novel PDH ( PDH ) that injects pairwise geometric context into coarse retrieval to improve accuracy. In the refinement stage, DBH -based filtering eliminates attribute-inconsistent correspondences, and yaw voting using matched triangle relationships rejects pose-inconsistent matches and suppresses outliers caused by hash collisions. These components reduce false positives from structural aliasing and preserve pose-consistent inliers, resulting in robust candidate correspondences for subsequent verification. Given reliable candidates, TreeLoc++ performs geometric verification and fine-grained pose estimation through two complementary strategies. It evaluates alignment quality using an overlap score that incorporates a spatial proximity prior and discourages associations between geometrically similar but spatially distant scenes. This design reflects the fact that stem reconstruction quality is viewpoint-dependent, and shared trees between distant scenes are often observed under different conditions, leading to inconsistent geometry and increased mismatches. Additionally, TreeLoc++ applies a constrained 6-DoF optimization that jointly estimates roll, pitch, and height by enforcing geometric consistency among matched tree geometry, in contrast to prior methods that estimate these components independently. As shown in Fig. 1 , TreeLoc++ offers a lightweight and scalable alternative to conventional point cloud-based localization systems.