Pose correction and efficient grasping based on non-contact sensing for robotic apple picking

Achieving an optimal grasping pose is essential for stable robotic apple picking. However, vision-based localization may become unreliable during the final approach because of foliage occlusion, depth errors, and limited visibility. Without real-time near-field feedback, the gripper cannot reliably correct fruit–gripper misalignment before contact, increasing the risks of asynchronous contact, slippage, and mechanical damage. To address this issue, this study proposes a non-contact proximity perception method based on electrostatic induction between the apple and the sensor. Three ionogel-based triboelectric nanogenerator (IG-TENG) sensors are mounted on a three-finger Fin-Ray flexible gripper. By exploiting the differential non-contact responses of the three sensors, a pose-correction algorithm based on normalized signal weights is developed to achieve dynamic alignment between the apple and the gripper without requiring an absolute voltage-to-distance calibration model. Furthermore, a parallel control strategy leveraging the IG-TENG sensors' dual-mode non-contact and tactile sensing capabilities enables real-time trajectory correction and an automatic transition to force-controlled grasping upon contact. Laboratory experiments achieved a mean residual planar error of 2.62 ± 1.05 mm and a pose-correction success rate of 100%. Orchard trials achieved a picking success rate of 89.0% with an average cycle time of 6.9 s per fruit. Compared with the conventional vision-only system, the picking success rate increased by 16 percentage points, and the grasping time was reduced by 64%. These findings demonstrate the practical feasibility of IG-TENG-enabled near-field sensing for improving grasping accuracy, harvesting efficiency, and operational reliability in complex orchard environments.