Haptic Object Parameter Estimation during Within-Hand- Manipulation with a Simple Robot Gripper

Though it is common for robots to rely on vision for object feature estimation, there are environments where optical sensing performs poorly, due to occlusion, poor lighting or limited space for camera placement. Haptic sensing in robotics has a long history, but few approaches have combined this with within-hand-manipulation (WIHM), in order to expose more features of an object to the tactile sensing elements of the hand. As in the human hand, these sensing structures are generally non-homogenous in their coverage of a gripper's manipulation surfaces, as the sensitivity of some hand or finger regions is often different to other regions. In this work we use a modified version of the recently developed 2-finger Model VF (variable friction) robot gripper to acquire tactile information while rolling objects within the robot's grasp. This new gripper has one high-friction passive finger surface and one high-friction tactile sensing surface, equipped with 12 low-cost barometric force sensors encased in urethane. We have developed algorithms that use the data generated during these rolling actions to determine parametric aspects of the object under manipulation. Namely, two parameters are currently determined 1) the location of an object within the grasp 2) the object's shape (from three alternatives). The algorithms were first developed on a static test rig with passive object rolling and later evaluated with the robot gripper platform using active WIHM, which introduced artifacts into the data. With an object set consisting of 3 shapes and 5 sizes, an overall shape estimation accuracy was achieved of 88% and 78% for the test rig and hand respectively. Location estimation, of each object's centroid during motion, achieved a mean error of less than 2mm, along the 95mm length of the tactile sensing finger.