GPS-based navigation in orchards can be unstable because trees may block the GPS signal or introduce multipath errors. Most research on robot navigation without GPS has focused on guidance inside orchard rows; end-of-row detection has not received enough attention. Additionally, navigation between rows relies on reference maps or artificial landmarks. In this work, a novel row-end detection method is presented, which detects drastic changes in the statistical distribution of the sensed point cloud as the robot gets closer to the row’s end. A row-entry method was also implemented that builds a local map that is used by a reactive path tracker. The system was evaluated in a 24-row block in a vineyard. Once the robot was closer than 7 m from the end of a row, the algorithm detected it with a 100% success rate and calculated the distance from it with a mean error of 0.54 m. The system was also evaluated in vineyard configurations with parallel and slanted vine rows in consecutive blocks. The system worked well in all configurations, except where the next block had rows aligned to the rows of the current block and the headland width was closer than 5 m.

Strawberries naturally grow in clusters, interwoven with leaves, stems, and other fruits, which frequently leads to occlusion. This inherent growth habit presents a significant challenge for robotic picking, as traditional percept-plan-control systems struggle to reach fruits amid the clutter. Effectively picking an occluded strawberry demands dexterous manipulation to carefully bypass or gently move the surrounding soft objects and precisely access the ideal picking point—located at the stem just above the calyx. To address this challenge, we introduce a strawberry-picking robotic system that learns from human demonstrations. Our system features a 4-DoF SCARA arm paired with a human teleoperation interface for efficient data collection. We propose the end pose assisted action chunking transformer (EPACT) to learn a fine-grained visuomotor policy. By incorporating end-pose supervision, our approach enables the robot to acquire dexterous manipulation skills such as pushing aside leaves or navigating tight clusters that are difficult to achieve with traditional pipelines. Experiments demonstrate that our method successfully transfers human-like dexterity to the robot, enabling clustered strawberry picking and outperforming imitation learning baselines, underscoring its potential for practical application in complex agricultural environments.