Localization and Mapping on Agriculture Based on Point-Feature Extraction and Semiplanes Segmentation From 3D LiDAR...

Area of Science:

• Robotics
• Computer Vision
• Agricultural Engineering

Background:

• Agricultural robotics requires robust localization and mapping for tasks like spraying and harvesting.
• Unstructured agricultural environments pose significant challenges for existing localization algorithms.
• Advancements in localization techniques are crucial for the widespread adoption of agricultural robots.

Purpose of the Study:

• To introduce VineSLAM, a novel algorithm for precise localization and mapping in agricultural settings.
• To address the challenges posed by unstructured environments in agricultural robotics.
• To improve the navigation capabilities of ground robots in vineyards.

Main Methods:

• VineSLAM utilizes point- and semiplane-features extracted from 3D LiDAR data.
• A novel Particle Filter integrates both feature modalities for enhanced localization.
• The algorithm's numerical stability was validated using simulated data.

Main Results:

• VineSLAM demonstrated accurate robot localization using only three orthogonal semiplanes.
• Real-world experiments in woody-crop vineyards showed VineSLAM's superior performance.
• The algorithm achieved high precision in long, symmetric vineyard corridors, outperforming state-of-the-art methods.

Conclusions:

• VineSLAM offers a robust solution for localization and mapping in agriculture.
• The proposed algorithm significantly improves robot navigation precision in challenging vineyard environments.
• VineSLAM is a key advancement for enabling autonomous operations in agricultural robotics.