The SLS-Generated Soft Robotic Hand - An Integrated Approach Using Additive Manufacturing and Reinforcement Learning

To develop a robotic system for a complex task is a time-consuming process. Merging methods available today, a new approach for a faster realization of a multi-finger soft robotic hand is presented here. This paper introduces a robotic hand with four fingers and 12 Degrees of Freedom (DoFs) using bellow actuators. The hand is generated via Selective Laser Sintering (SLS), an Additive Manufacturing method. The complex task execution of a specific action, i.e. the lifting, rotating and precise positioning of a handling-object with this robotic hand, is used to structure the whole development process. To validate reliable functionality of the hand from the beginning, each development stage is SLS-generated and the targeted task execution is trained via Reinforcement Learning, a machine learning approach. Optimization points are subsequently derived and fed back into the hardware development. With this Concurrent Engineering strategy a fast development of this robotic hand is possible. The paper outlines the relevant key strategies and gives insight into the design process. At the end, the final hand with its capabilities is presented and discussed.