Intelligent automated gesture planning for reconfigurable soft gripper system in food handling

Soft grippers, that can mimic human fingers grasping objects, have emerged as a game-changer in the automated food service industry. Their flexibility and compliance enable them to handle various food items, regardless of size, shape, or stiffness, outperforming their rigid counterparts while maintaining cost-efficiency and greater adaptability. However, the working scenarios for soft grippers are also more complex and unpredictable, leading to a challenge in pre-programming the gesture and trajectory of the system. Current research on soft grippers primarily focuses on their compliance characteristics, which is effective for objects with characteristic features but faces challenges with cut food items. For such cut objects, both vertical friction and compliance play crucial roles in grasping, highlighting the need for a reconfigurable soft gripper system (RSGS) with multiple degrees of freedom actuators. To address these challenges, we introduce an intelligent automated gesture planning strategy for RSGSs with multiple degrees of freedom. Our proposed framework comprises five modules: Feature Engineering, which parameterizes and samples arbitrary polygon cross-sections of food items; Simulation, which automates the creation of numerical models in the design space, run and post-processing of simulation; GraspingFormer, which estimates reaction forces during grasping; AgentVAE, which uses a generative variational autoencoder to sample feasible grasping gestures in latent space; Planner, which identifies the optimized gestures by solving an inverse problem. This strategy can facilitate gesture planning when grasping a target object with a RSGS, to enhance the picking-up success rate. The proposed framework can potentially benefit food-handling-like tasks and expand the use of soft robots in real-world applications.