An efficient data-driven framework of hybrid dynamics for real-time modeling and control of continuum robots

Continuum robots exhibit exceptional compliance and adaptability. However, their intrinsic nonlinear dynamics poses significant challenges for real-time control under dynamic conditions. This paper presents a new data-driven framework of hybrid dynamics (HD) for modeling and control of continuum robots. This framework decomposes the nonlinear dynamics of a continuum robot into the nonlinear statics of global deformation and the linear dynamics of relative vibration, and leads to two sub-models of global deformation statics (GDS) and relative motion dynamics (RMD) via different data-driven approaches. The first sub-model uses a deep neural network to predict nonlinear static deformations, while the second sub-model utilizes dynamics mode decomposition for relative vibration compensation. The nonlinear integration of the two sub-models establishes the physics-embedded data-driven model with reduced complexity. Combining quasi-static positioning and model-predictive vibration suppression leads to design of a hierarchical dynamic controller. The experiments of a cable-driven continuum robot demonstrate precise trajectory tracking and effective vibration suppression under payload and high-speed conditions. This data-driven modeling and control framework balances computational efficiency and control performance, enabling practical advances of continuum robots under complex working scenarios.