Proceedings of the
9th International Conference of Asian Society for Precision Engineering and Nanotechnology (ASPEN2022)
15 – 18 November 2022, Singapore

Gesture Selection Strategy of Reconfigurable Soft Gripper System Using a Data-Driven Scheme

Yangfan Li1, Jun Liu1, Zhuangjian Liu1,a, Jin Huat Low2, Jin Jin2 and Chen Hua Yeow2

1Institute of High Performance Computing, A*STAR, 1 Fusionopolis Way, 138632, Singapore

2Advanced Robotics Centre, National University of Singapore, 117608, Singapore


Soft grippers, designed to mimic human fingers, have shown great potential in the automated food handling industry. Due to their highly compliant and flexible characteristics, they could handle rigid or delicate soft food samples of different shapes and sizes, superior to their rigid-bodied counterparts with a lower manufacturing cost and higher flexibility. Individually reconfigurable soft gripper systems, consisting of three or more fingered grippers, has been developed, which could further boost the efficiency and productivity as its dynamic gripping range has been greatly improved. However, the soft nature of gripper and food object also significantly increases the operation complexity. In real application, we generally cannot pre-program all the situations due to the complex scenario in the food pick-and-place task. Therefore, how to automatically adjust the fingers to pick and place an item with the best chance of success becomes a main concern in the food industry application. To address this challenge, this work proposed a data-driven decision strategy to determine the optimized gesture of the reconfigurable gripper system to handle a food object. Firstly, we have established an efficient simulation model to predict the mechanical behavior during grasping, and extract useful information such as reaction forces, with reasonable accuracy. The target items to grasp have been simplified as quadrilaterals, and then parametrized within a fair range. The grasping gestures have also been parameterized within the dynamic operation range of the soft grippers. Then we used the simulation model to generate a large database for different combinations of gripping gesture and objects of various shapes and sizes. Subsequently, we built a surrogate artificial neural network model that can make fast prediction of the mechanical performance of the soft gripper system on a given item of arbitrary design parameters. Finally, this surrogate model was used in an optimization algorithm to obtain the optimized grasping gesture, aiming to enhance the success ratio for picking up the target object. With the help of an object detector, the established model and strategy is expected to provide real-time advice on the optimized gripping gesture to achieve the best grasping success ratio.

Keywords: Data-driven, Reconfigurable soft gripper, Machine Learning, Soft robotics, Automation.

PDF Download