New framework enables agricultural robots to detect vine trunks in vineyards using limited labeled data and multi-sensor fusion

Researchers developed an annotation-to-detection framework that trains robust multi-modal detectors with limited and partially labeled training data

The system uses cross-modal annotation transfer and early-stage sensor fusion to enhance detection capabilities without requiring large-scale manually labeled datasets

Framework demonstrates effectiveness for vine trunk localization in novel vineyard environments with diverse, unstructured conditions

Addresses key challenge of deploying autonomous mobile robots in dynamic agricultural fields where real-time detection is critical