<p>Against the backdrop of global agricultural automation and labor shortages, intelligent fruit harvesting and sorting have become critical requirements. However, existing systems still suffer from issues such as inaccurate fruit ripeness identification, insufficient grasping flexibility, and inadequate integrated perception capabilities in grippers. However, existing systems often face challenges such as inaccurate ripeness identification, insufficient grasping flexibility, and a lack of integrated perceptual capabilities in grippers. To address these limitations, this study presents a novel three-finger, six-segment rigid sorting end-effector with a fin-ray structure. For litchi handling, a flexible perception gripper incorporating variable-size mechanisms and hydrogel pressure sensors is employed. This gripper is integrated with a machine learning-based classification system to evaluate fruit ripeness and freshness, enabling the adaptive grasping of objects with diverse dimensions and rigidity. Four time-series models—1D-CDD, TCN, LSTM, and GRU—were evaluated via 5 × 5 cross-validation. The TCN model demonstrated superior performance, achieving an average accuracy of 92% in maturity and freshness recognition while exhibiting high stability. This research confirms that a tactile gripper equipped with variable-size attachments can fulfill the flexible grasping demands of robotic sorting. When combined with machine learning algorithms, the system can accurately estimate litchi ripeness and freshness, provides a novel technical pathway and preliminary theoretical and practical foundation for future development of intelligent gripping and sorting systems applicable to real-world scenarios.</p>