Towards a Universal Modeling and Control Framework for Soft Robots

Abstract

Traditional rigid-bodied robots are designed for speed, precision, and repeatability. These traits make them well suited for highly structured industrial environments, but poorly suited for the unstructured environments in which humans typically operate.

Soft robots are well suited for unstructured human environments because they them to can safely interact with delicate objects, absorb impacts without damage, and passively adapt their shape to their surroundings. This makes them ideal for applications that require safe robot-human interaction, but also presents modeling and control challenges. Unlike rigid-bodied robots, soft robots exhibit continuous deformation and coupling between structure and actuation and these behaviors are not readily captured by traditional robot modeling and control techniques except under restrictive simplifying assumptions.

The contribution of this work is a modeling and control framework tailored specifically to soft robots. It consists of two distinct modeling approaches. The first is a physics-based static modeling approach for systems of fluid-driven actuators. This approach leverages geometric relationships and conservation of energy to derive models that are simple and accurate enough to inform the design of soft robots, but not accurate enough to inform their control. The second is a data-driven dynamical modeling approach for arbitrary (soft) robotic systems. This approach leverages Koopman operator theory to construct models that are accurate and computationally efficient enough to be integrated into closed-loop optimal control schemes.

The proposed framework is applied to several real-world soft robotic systems, enabling the successful completion of control tasks such as trajectory following and manipulating objects of unknown mass. Since the framework is not robot specific, it has the potential to become the dominant paradigm for the modeling and control of soft robots and lead to their more widespread adoption.