Modeling and System Identification of Pneumatic Robot

Abstract

Pneumatic actuators can prove to be really useful in robotics as they have natural backdrivability, good strength to weight ratio and lot of other interesting properties. However, they are quite difficult to control owing to their non linearity and high latency due to air’s compressability. One of the promising ways to make these actuators usable in advanced robotic tasks is to use model based control. Thus, this work is entirely focused to obtain a model for pneumatic actuators which has good predictability for different kinds of control objectives. In this work, first a physics model is derived through first principles and it’s parameters are identified through classical regression techniques and carefully conducted experiments. Then in order to obtain even higher accuracy, neural network based data driven pure and hybrid models are proposed. Finally, all of these models are validated and evaluated by comparing their predictions to the actual hardware data, which is as diversified as possible.