A hand-like gripper embedded with flexible gel sensor for tomato harvesting: soft contact and intelligent ripeness sensing

With the continuous growth of tomato yields and the need to address the cumbersome task of manual harvesting, the development of harvesting robots presents a promising solution for tomato harvesting. A key challenge lies in improving the success rate of these robots for reliable tomato harvesting without causing damage. This study aims to propose a rigid-flexible coupled gripper to ensure higher success rates in tomato harvesting, while also incorporating an advanced method for detecting fruit ripeness. The paper introduces a flexible hydrogel pressure sensor, featuring a wide detection range, high sensitivity, and excellent stability, integrated into the gripper design. A signal acquisition system based on a sensor array is developed, enabling the accurate capture of force signals during the tomato grasping process. The sensor array collects tactile sequence data in real-time, which, when combined with compressive deformation data from the fruit, forms a comprehensive dataset. To detect tomato ripeness, four classification models—Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), Multi-Layer Perceptron (MLP), and Fully Convolutional Networks (FCN)—are implemented. Among the four models, the LSTM network achieves the highest classification performance. The overall recognition accuracy for tomatoes of different ripeness levels, within the same variety and at room temperature, is determined to be 99%. The results demonstrate that the combination of a flexible gripper with hydrogel sensors not only preserves the integrity of the tomatoes but also ensures accurate detection of fruit ripeness. This innovation has the potential to significantly enhance the performance of tomato harvesting robots, contributing to more efficient and automated agricultural practices.