Kinematic control of soft robotic manipulators is a challenging problem particularly for systems that are both globally and locally redundant. This article presents a learning-based task-space kinematic controller that enables tracking in such soft robotic manipulators. The novelty of the work is a bioinspired structured sampling mechanism that actively regulates the variance in motor movements during motor exploration. It generates a database that is applied to a direct learning architecture, thereby, formulating an inverse model at the position-level. The controller is validated in simulation on a 12 degrees-of-freedom modular manipulator comprised of elemental modules with three longitudinal actuators and one radial actuator. Experiments demonstrate consistency in performance across multiple unseen trajectories and repeatability of each task. Furthermore, the performance remains uncompromised in altered motor conditions, provided task-relevant motor variance. The results exhibit accurate, repeatable, and adaptive tracking behavior of the system and are promising for the advancement of these systems.
Structured motor exploration for adaptive learning-based tracking in soft robotic manipulators
Ansari Y.;Laschi C.;Falotico E.
2019-01-01
Abstract
Kinematic control of soft robotic manipulators is a challenging problem particularly for systems that are both globally and locally redundant. This article presents a learning-based task-space kinematic controller that enables tracking in such soft robotic manipulators. The novelty of the work is a bioinspired structured sampling mechanism that actively regulates the variance in motor movements during motor exploration. It generates a database that is applied to a direct learning architecture, thereby, formulating an inverse model at the position-level. The controller is validated in simulation on a 12 degrees-of-freedom modular manipulator comprised of elemental modules with three longitudinal actuators and one radial actuator. Experiments demonstrate consistency in performance across multiple unseen trajectories and repeatability of each task. Furthermore, the performance remains uncompromised in altered motor conditions, provided task-relevant motor variance. The results exhibit accurate, repeatable, and adaptive tracking behavior of the system and are promising for the advancement of these systems.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.