Data-Driven Methods for Sensing, Modeling and Control of Soft Continuum Robot: A Review

Abstract

Inspired by soft-bodied animals, soft continuum robots provide inherently safe and adaptive solutions in robotics, especially suited for applications requiring gentle interactions. However, challenges arise from the nonlinearity and hysteresis of soft materials, coupled with their infinite degrees of freedom, complicating the sensing, modeling, and control of these robots. Data-driven methods, which leverage observations of system behavior, present a promising approach for predicting the dynamic properties of soft continuum robots. This article reviews the advancements in data-driven sensing methods for self-configuration feedback and environmental perception. We explore various data-driven techniques for both kinematic and dynamic modeling, and we discuss data-driven control methods, including supervised learning, and reinforcement learning. We envision that empowering soft continuum robotic systems with data-driven approaches may facilitate a broad range of applications, such as executing dynamic tasks, adapting to random disturbances, and functioning effectively in spatially and temporally varying environments.