Torque-Sensing Soft Bellows Actuator for Multi-Fingered Hands Taking Bellow's Buckling Into Consideration

Abstract

Soft multi-fingered robotic hands are safe in human environments and can perform human-like behaviors. However, structural complexities and nonlinearities in soft actuators complicate torque sensing, a critical function for dexterous object manipulation. This study introduces a torque-sensing finger joint mechanism using a soft actuator composed of water-powered hydraulic bellows. Two real-time torque estimation methods are proposed, developed, and validated specifically for cases in which buckling occurs in the bellows, a situation that typically presents significant estimation challenges. The buckling spring model for torque estimation, explicitly considering the buckling effect, is a linear model that considers two elastic forces for the pressure and external force. The multi-layer perceptron model for torque estimation considers the nonlinearity of the actuator. The experimental results show that both methods can estimate the torque in real-time with high accuracy. The torque control for grasping fragile objects has also been examined in real-world scenarios. The findings indicate that, compared to the approach without torque control, successful and safe manipulation of the target objects is accomplished without causing detrimental deformation.