Efficient Ripeness Monitoring in Open-Facility Environments Using a Quadruped Robot and Panoramic AI Recognition

Abstract

Efficient facility-scale tomato ripeness monitoring remains difficult in greenhouses where uneven terrain limits conventional wheeled and rail-guided platforms and planar cameras provide restricted coverage. This study developed a wheel-legged quadruped monitoring system integrating LiDAR, a depth camera, and a panoramic camera. An adaptive gait-switching strategy supported navigation across heterogeneous terrain. Panoramic images were projected into six perspective views, and the left and right views were processed using a YOLOv8-based ripeness recognition model. Time-synchronized detections and robot poses were fused to map ripeness observations into three-dimensional greenhouse coordinates. Five field experiments in a commercial tomato facility demonstrated autonomous row traversal, inter-row transition, and avoidance of pedestrians, obstacles, and cultivation boundaries. The recognition pipeline continuously identified multiple ripeness stages under variable illumination, foliage occlusion, and robot motion, while the spatial fusion procedure produced a facility-scale three-dimensional ripeness distribution. The integration of terrain-adaptive quadruped mobility, panoramic perception, and spatial mapping provides a practical framework for continuous ripeness monitoring and can support targeted harvesting, yield forecasting, and crop management.

Plant phenotyping relevance

トマト果実の成熟度を画像認識で取得するロボット型フェノタイピングシステムを開発し、実環境で評価しており、表現型取得法が研究の中心である。

Code and data availability

公開論文であることは確認できましたが、現在の公式API・許可済み取得経路では本文を自動取得できませんでした。

No evidence-backed public reproduction asset is currently recorded.