Haptic based surface stiffness identification using machine learning

Abstract

Touch is an essential environmental input for all living things, including humans. Most of the objects with which individuals interact are soft and malleable. Humans have inherited the ability to perceive and recognize differences in the deformable features of objects. Robotic systems have been developed to support numerous industries, and in robotics, “haptics” refers to forces and force feedback from the object. In many sectors, robotic devices handle deformable objects. We believe that to improve operational quality, robotic systems must be able to recognize deformable objects. While it is possible to observe the characteristics of items during their manipulation, haptic-based object identification remains a challenging subject due to the complexity of deformable object characteristics. The discussion of object classification methodology begins with collecting object deformation data and sensors for data collection. Sensor array-based measurement techniques are capable of observing the pressure variation of the deformation area, while single point measurement techniques are capable of observing the force variation and compression depth of the object. The use of force response and compression distance measurements enables the extraction of additional attributes of deforming objects, such as stiffness, hysteresis, velocity, acceleration fluctuation, and energy absorbed during compression. The use of machine learning to classify objects with features avoids many disadvantages associated with traditional mathematical model-based classification methodologies. The ability to handle time series data and large amounts of data are also key features of machine learning. In this study, we introduce the construction of additional features that may improve classification and use the Time Series Forest Classifier (TSFC) and permutation importance to identify the best performing features for object classification. Keywords: Haptic, Haptics object modeling, deformable objects, Haptics features for machine learning, disturbance observer, reaction force observer, time series classification