Model-Free Intelligent Control for Space Soft Robotic Manipulators

Abstract

Given the advantages of softness, lightness, low cost, and interaction safety, inverse kinematic modeling and control of soft actuators has caused a research boom. However, in realizing dexterous manipulation of space large soft manipulators, it is much more difficult to achieve precise control not only because of the greater accumulation of errors in the multiple degrees of freedom and nonlinear properties of soft materials at large scales but also because of the inability of directly solving the inverse kinematics in the cases of singular pure elongation. In this work, a model-free intelligent kinematic control strategy is proposed for these problems that exhibit a mapping relationship between the output end-effector position and the input pressure. For multiple-degree-of-freedom robots, especially pneumatic soft manipulators, traditional inverse kinematic modeling methods are complex and inverse Jacobian matrix solution often encounters geometric singularities. To address this issue, this paper proposes an inverse kinematics–multilayer perceptron (IK-MLP) method for soft manipulators. In this strategy, the trained intelligent controller can be applied to control pneumatic manipulators without establishing a traditional inverse kinematic model. The control algorithm is experimentally tested based on the ground experiment system of the space soft manipulator. Simulations and experiments are carried out to validate the given model-free intelligent controller, proving that the IK-MLP method can effectively solve the singularity of inverse kinematics.