Two-fingered, tactile-based manipulation of unknown objects

Modern sensor arrays providing tactile feedback with high spatial and temporal resolution allow for new control strategies to exploit this valuable sensory channel for grasping and manipulation tasks. We present our recent efforts to transfer a control approach specifically developed to exploit tactile information for inhand manipulation of objects [3] from simulation to the real world. Aiming to handle unknown objects, the control strategy makes as parsimonious assumptions about available prior knowledge as possible: Neither the object properties (shape, weight) nor contact properties (friction coefficients, softness) are assumed to be available. Rather, we assume that the contact position, normal direction and contact force is known from a tactile sensor array providing sufficient spatial and temporal resolution. Because an appropriate tactile fingertip sensor so far only exists as a prototype [1] and needs to be integrated into our robot hand, we employ a previously developed, tactile sensor array [6] mounted as large fingertips on two Kuka LWR arms (see Fig. 1) to demonstrate the feasibility of the approach in the real world. The tactile sensor array of 16×16 tactile elements, sized 5×5mm each, provides a frame rate of 1.9kHz and a linear measurement range between 4-17kPa rendering its use for real-time robot control feasible. Employing image processing methods, we compute the contact location on the sensor as the weighted center of forces within the largest connected contact region. Exploiting the known kinematics of the robot we can transform these sensor-local coordinates into the robot’s reference frame. From the known shape of the tactile sensor we extract the contact normal direction thus providing