Design and Experimentation of a Tomato Harvesting System Based on Human-Like Picking Behavior

Addressing the challenges of tomato harvesting, such as fruit susceptibility to damage, incomplete positional information, and prolonged harvesting time, this study explores human-like artificial tomato picking behavior. It involves designing a mechanical hand that mimics human hand-picking methods, developing a tomato pose recognition system, proposing an efficient harvesting strategy, and conducting systematic integration and experimental verification. Results indicate that the mechanical hand can grasp a mass exceeding 1.4 kg, achieving an average tomato coverage of 3.5–6.5%, and enabling stable, damage-free grasping of tomatoes with diameters ranging from 60 to 105 mm. The success rate of stem recognition by the vision system reaches 85.23%, with an overall success rate of 82.42%. Absolute error between camera positioning and actual distance remains below 16mm, with relative error below 3.2%. After incorporating tomato pose information into the harvesting system, the lateral gripping method outperforms the bottom gripping method in terms of speed, interference resistance, and success rate. The average single-operation time ranges from 9.52 to 10.43 s, with success rates of 86.8–93.1%, achieving the objective of damage-free, precise, and efficient tomato harvesting.