CoChat

Precise robot localisation for vineyard pruning operations.

Stenger, Philip · 2026 · University of Canterbury

The DOI matches its DataCite record; the references it cites are not checked yet.

Abstract

Autonomous vineyard pruning demands low-centimetre-level localisation to guide a robotic arm to individual canes, yet GNSS-degraded row-crop corridors with repetitive geometric structure present fundamental challenges for conventional pose estimation. Even well-tuned LiDARinertial odometry accumulates decimetres of drift in such environments, an order of magnitude too coarse for selective cane removal. The operational workflow further complicates the problem: a high-fidelity 3D Gaussian Splatting model is constructed offline from a prior vineyard visit, and the robot must re-localise against this pre-built reference on subsequent passes. This offlinemodel re-alignment scenario remains under-addressed by existing vineyard pruning approaches, which predominantly construct reference representations in real-time. This thesis presents a two-stage hierarchical localisation architecture for this domain. The first stage employs a LiDAR-inertial SLAM system augmented with sparse GNSS factor injection (GLIM-GNSS) to produce a low-decimetre global pose estimate. The second stage refines this estimate through Generalised Iterative Closest Point registration of live sensor data against the pre-built 3D Gaussian Splatting model. Both LiDAR and stereo vision are evaluated as sensing modalities for the refinement stage across a factorial experimental design varying sensor-to-vine distance and approach angle. The SLAM system achieves a mean absolute trajectory error of 103mm across five vineyard/ orchard datasets, corresponding to a mean error rate of 0.022% per metre traversed, within the low-decimetre accuracy required for downstream registration. For precise alignment, the LiDAR pipeline achieves 100% condition-level convergence with a median inlier RMSE of 33.25 mm, demonstrating that reliable low-centimetre registration is achievable across all tested operating conditions. The stereo pipeline attains the lowest registration errors, with a mean inlier RMSE of 15.51mm and performance reaching 12mm at 80 cm operating distance, achieving the accuracy required for direct manipulator guidance. The applied Iterative Closest Point corrections (47–165mm) fall within the SLAM system’s error bounds, confirming that the coarse pose reliably places the robot within the registration algorithm’s convergence basin. These results validate the two-stage design premise for vine-scale localisation. The architecture achieves low-centimetre pose accuracy using sensors already present on pruning platforms, with LiDAR providing reliable convergence across all tested conditions and stereo achieving sub-15mm registration precision suitable for pruning-scale manipulation.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

DataCite · 10.26021/16730

Cites

Statistics

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Metadata

Authors
Stenger, P.
Venue
University of Canterbury
Publisher
University of Canterbury
Published
2026

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Precise robot localisation for vineyard pruning operations. https://app.cochat.ai/papers/10.26021/16730
The paper
Stenger, P. (2026). Precise robot localisation for vineyard pruning operations. University of Canterbury. https://doi.org/10.26021/16730

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.