Abstract

Object recognition using the tactile sense is one of the leading human capacities. This capability is not as developed in robotics as other sensory abilities, for example visual recognition. In addition to a robot???s ability to grasp objects without damaging them, it is also helpful to provide these machines with the ability to recognise objects while gently manipulating them, as humans do in the absence of or complementary to other senses. Advances in sensory technology have allowed for the accurate detection of different types of environment; however, the challenge of being able to efficiently represent sensory information persists. In this paper, a sensory system is proposed that allows a robotic gripper armed with pressure sensors to recognise objects through tactile manipulation. A pressure descriptor is designed to characterise the voltage magnitudes across different objects and, finally, machine learning algorithms are used to recognise each object category. The results show that the pressure descriptor characterises the different classes of objects in this experimental set-up. This system can complement other sensory data to perform different tasks in a robotic environment and future research areas are proposed to handle problems with tactile manipulation.