The International Journal of Robotics Research Study, Ahead of Publish.
Soft robotics are testing to version and control because of their improperly specified kinematics and nonlinear characteristics. Lately, Koopman driver concept has actually been revealed with the ability of creating control-oriented soft robotic versions from information. Nonetheless, constructing these versions calls for comprehensive information collection and they do not always generalise well beyond the training monitorings. This paper provides an extra data-efficient and generalizable technique to soft robotic modeling that initially determines a physics-based Koopman version after that supplements it with a data-driven recurring Koopman version. The resulting integrated version is direct and therefore suitable with real-time model-based control methods such as Design Predictive Control (MPC). The effectiveness of the technique is shown on a number of substitute systems and on an actual soft robotic arm, where it is revealed to produce versions that are a lot more precise than totally physics-based versions and call for much less information to construct than totally data-driven versions. Making use of a model-based controller, the soft arm has the ability to effectively track end effect trajectories, execute a pick-and-place job, and create on a dry-erase board, showcasing the applicability of this structure to boost the abilities of soft robot systems.
发布者：Daniel Bruder，转转请注明出处：https://robotalks.cn/a-koopman-based-residual-modeling-approach-for-the-control-of-a-soft-robot-arm/
