Pressure-insensitive flexible bending sensor-enabled soft gripper for machine learning-based tomato size-sorting

Flexible bending sensors can significantly enhance soft grippers in robotic tomato sorting systems by equipping them with the precise detection capabilities needed for gentle and accurate handling. However, conventional sensors are susceptible to contact pressure interference during grasping, leading to distorted bending signals and reduced sorting precision. To overcome this limitation, this paper presents a novel fully flexible bending sensor with pressure-insensitive characteristics. The sensor combines a hemispherical bulge structure with an H-shaped microchannel design, exhibiting exceptional resistance to pressure interference. Experimental results show the sensor achieves a bending-to-pressure sensitivity ratio of 47.5:1, a linearity of R 2 = 0.97, a response time of 144 ms, and excellent cycling stability. The sensor was integrated into a flexible gripper, where a regression forest model established the correlation between grasping resistance values and tomato sizes, enabling highly accurate size prediction. Finally, this pressure-insensitive, high-sensitivity flexible bending sensor was successfully implemented on a robotic arm for automated sorting, making a significant contribution to intelligent agricultural equipment development.