Answers are grounded in this paper and cite their sources.
Greenhouse robots must perform navigation and fruit sensing under irregular canopy boundaries, occlusion, changing observation conditions, and constrained row-transition spaces. This study developed a simulation-driven ROS 2–Gazebo workflow for evaluating an omnidirectional mobile robot integrating LiDAR-based crop-row navigation, AprilTag-guided multi-row mission control, and ripeness-specific cherry-tomato counting. The system combined LiDAR row centering, PTZ-RGB and front RGB cameras, AprilTag-based mission control, YOLO instance segmentation, ByteTrack, and canonical-identity-based counting. The workflow was evaluated in a procedurally generated greenhouse with heterogeneous plant structures and known plant-level fruit distributions. The AprilTag-guided mission reproduced the intended multi-row route at commanded speeds of 0.2–1.0 m/s, with mean cross-track errors of 0.045–0.056 m and RMSE values below 0.10 m. Fruit-counting performance was strongly affected by robot speed. At 0.2 m/s, the system recovered 90.1% of the ground-truth fruit count with a count-level F1-score of 0.898. Performance declined above 0.4 m/s because of reduced observation time, canopy occlusion, and unstable temporal association. Additional errors arose from canopy obstruction and confusion between orange and red fruits. The proposed workflow enables repeatable evaluation of navigation, mission control, perception, and counting functions before physical deployment. It supports early identification of technical limitations and speed–accuracy trade-offs, thereby reducing trial-and-error, shortening the lead time to real-greenhouse experiments, and decreasing the number and duration of physical tests required for precision-agriculture technology development.
No takes yet. Share an insight, caveat, or question.
Awal et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: