Sim-to-Real Transferable Object Classification Through Touch-Based Continuum Manipulation

It is important to investigate object perception for classification or recognition based on touch sensing, especially when robots are operating in darkness or the objects are difficult to capture by vision sensors. In this work, we present a new form of continuum manipulator equipped with sparse touch sensing, validate the effectiveness of automatic generation of the touch-based continuum wraps, and the effectiveness of object classification based on the continuum wraps. Using the indirect object shape information encoded in the robot shape, we demonstrate that a classifier trained from the simulated continuum wraps is transferable to identify the real world objects with real continuum wraps.