Data-driven-based Stable Object Grasping for a Triple-fingered Under-actuated Robotic Hand

This paper proposes a data-driven-based stable object grasping method to grasp and realize object force/torque equilibrium for a triple-fingered under-actuated robotic system. Controlling the fingertip force of an under-actuated hand for precision grasp has been a challenging task in the robotic field due to the high nonlinearity of the under-actuated system as well as the inaccuracy in model-based controllers. By attaching a force sensor to the suitable location inside the robotic hand, necessary information for stable grasping, which is the detection of contact between fingertip and object and the magnitude of the fingertip force, is obtained using the proposed data-driven methods on internal sensor data in real-time. The obtained information is then used as feedback for the stable grasp controller to realize grasped object force/torque equilibrium.