Sensorless model-based object-detection applied on an underactuated adaptive hand enabling an impedance behavior

This paper introduces a sensorless object-detection control strategy designed for an underactuated three fingers hand. The proposed algorithm achieves a very accurate object-detection, through a simple joint impedance control scheme combined with a state-observer. The joint impedance control adapts the position set-point of each finger according to the estimated kinematics state computed from the kinematics model of the finger. The additional control loop is applied as external controller w.r.t. the standard one. The object-detection is obtained without any external sensors using only the measures of position and current provided by the fingers actuators. A friction compensation strategy based on a probabilistic approach has been implemented. An impedance control algorithm avoids the tuning of grasping parameters (i.e. grasping velocity and holding force). In this way developed controller provides adaptive behavior enabling the manipulation of objects made by unknown materials, without producing damages on their surfaces. This approach represents a step ahead to the flexibility of the grasping devices, in particular in manufacturing production, where the variability of manipulated pieces, in terms of shape and materials can be very high. An experimental campaign demonstrates the feasibility of the proposed method.