Introduction

Soft robots are widely applied in various areas, like surgery, human assistance and cooperation, wearable and rehabilitation devices, bionics, environment exploring, and grippers. As shown in Fig.1, modeling estimates robot states considering actuation variables and the natural properties of robots (shape, deformation properties, and actuation pattern.) Control decides the actuation variables with the current and target robot states, sensing signals, and the natural properties of robots. They can be seen as inverse processes, so we introduce modeling and control in parallel.

From the view of data models, soft robots have infinite degrees of freedom, which means such models generally require a large number of parameters. Soft robots provide nonlinear and delayed responses; such data models are complex.

Fig. 1. Diagram of soft robot modeling and control processes. Robot design (grey) provides a specific soft robot structure and actuation pattern. The sensing system (green) obtains robot information via the sensors of such a robot. Modeling f_θ is a function that predicts the robot state p (blue), as end position and robot pose, according to the actuation variables a (red.) Control g_τ aims to decide actuation variables a with the desired state p_d and sensing inputs. Finally, such a robot system can achieve a variety of tasks. θ and τ represent the parameters in the data-driven methods of modeling and control, respectively.