A vacuum-based robotic system for apple harvesting with local pose estimation and time-optimal path planning

Currently, apple harvesting remains heavily reliant on manual labor—a practice increasingly challenged by rising labor costs and a growing shortage of agricultural workers. Robotic harvesting offers a promising alternative due to its potential to reduce fruit damage and automate operations. However, issues such as low efficiency and high equipment costs continue to restrict its commercial viability. To address these limitations, this study presents a low-cost, vacuum-based robotic harvesting system integrating a vacuum end-effector, a 6 degree of freedom robotic arm, an Red-Green-Blue-Depth camera, and a mobile vehicle. The main innovations are threefold: (i) A novel estimation method combining an Improved You Only Look Once version 8 nano segmentation model with local point cloud features was developed to achieve fruit detection and orientation; (ii) a harvesting sequence optimization algorithm was introduced by modeling multi-target picking as a Traveling Salesman Problem and solving it with an improved Simulated Annealing algorithm that incorporates joint-space motion cost to minimize cycle time; and (iii) system integration and field validation. The prototype was tested in orchard environments, achieving a workspace coverage of 1.46 m × 0.82 m × 1.62 m, sufficient for most growing regions. Detachment tests established 46.7 N as the critical suction pressure required for stable pickup. In automated trials, the robot reached a 58.4% picking success rate, with an average cycle time of 6.36 s per fruit and a stem detachment rate of 17.6%. The prototype successfully demonstrated autonomous apple harvesting, revealing its considerable potential for practical deployment within complex orchard environments.