Data-Driven Predictive Disturbance Observer for Quasi Continuum Manipulators

Deriving accurate and time efficient models for soft robots has proven to be a difficult task. This is mainly due to the behaviour of the soft materials and the interaction with other parts of the robot, resulting in complicated models with many states that require complex sensor concepts.Soft robots such as the Bionic Soft Arm (BSA) are subject to uncertainties. The resulting mismatch between plant and model can be viewed as a disturbance. Disturbance observers significantly improve the control performance of the model-based controllers. However, they assume an underlying model of the disturbance that is usually unknown. To address this problem, we propose a novel Data-Driven Predictive Disturbance Observer (DPDO), which predicts the future disturbance based on past measurements that are updated online and, thereby, enables the model-based controller to achieve zero control error. The approach is first tested in simulation, examining the influence of noise, time delays and typical model uncertainties for the BSA. Afterwards, the algorithm is applied to the real system, demonstrating its practicability.