Reproduce “Tree-SLAM: semantic object SLAM for efficient mapping of individual trees in orchards” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
David Rapado-Rincon, Gert Kootstra
Tree-SLAM, developed by Wageningen University & Research, introduces a semantic SLAM framework specifically for orchards, accurately mapping individual trees in a 2D ground plane. The system integrates robust object detection, multi-stage data association, and a factor graph optimization to achieve tree geo-localization with a mean error as low as 0.18 meters for pear trees, outperforming a clustering baseline in most scenarios.
Key insights
- The effectiveness of PCA-based trunk center estimation is context-dependent, benefiting mature, well-defined trunks but potentially degrading accuracy for small, occluded trunks.
- Graph-based data association significantly improves mapping accuracy and robustness by utilizing the regular spatial arrangement of trees, especially in challenging visual conditions with occlusions.
- Integrating noisy multi-sensor data (GPS, odometry, semantic observations) within a factor graph effectively corrects for individual sensor inaccuracies, leading to more consistent and complete maps than simpler clustering methods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.