Abstract
There is an increasing interest to advance the state-of-the-art of soft robots due to its capability to extend the limits of traditional rigid robot manipulators. Many novel soft robots have been designed and proposed in the recent years of which, inflatable soft robots attract considerable research efforts, as the actuation mechanism is read- ily compatible with bio-inspired designs. However, the research on inflatable soft robots is currently at an early stage. Most prototypes lack a proper mathematical model and control design for real-time application. This study proposes a unified model identification framework for the pneumatic bending soft actuator and similar flexible systems. This method is one of the first attempts to deal with the complex nonlinear dynamics of soft robots that is challenging to model. The outcomes of this research could improve the current modelling and control design of inflatable soft robots for more advanced medical applications. Firstly, a statistics-based model identification approach is investigated. It introduces a newly developed model identification model structure, namely the Difference In- put Outputs PieceWise Linear Orthonormal Basis Function (DIO-PWL-OBF). This method collects local dynamic responses and linearises local subsystems that are approximated by a set of optimally selected OBFs. With the DIO setting, the switching between different subsystems is smooth especially in the input direction-dependent situations. The advantage is that this method can capture the local dynamics accurately and the local models can be used to construct global dynamics without adding offset terms or nonlinear weighting terms. The identification approach by the DIO-PWL-OBF model structure has been experimentally validated for common bending profiles using a three-chambered soft pneumatic actuator made of silicone. Regarding an identified DIO-PWL-OBF model structure, a solution that further improves its local partition and local linear approximations is presented. The anal- ysis shows that, under the DIO setting, the switching between two subsystems is coupling invariant whether in the same input direction or not. Utilising this unique feature, an iterative refinement algorithm is developed. The algorithm consists of two main steps, adjusting the offsets on individuals DIO sequences and redoing the linear approximate under eigenvalue constraints. The refinement algorithm has been experimentally validated by the soft actuator in a typical bending case.