LAUSR

Perceiving and identifying material properties of surfaces and objects is a fundamental aspect, which enables us to interact with a world. Automatic material identification, therefore, plays a critical role in intelligent manufacturing systems. In many scenarios, tactile samples and the tactile adjective descriptions about some materials can be provided. How to exploit their relation is a challenging problem. In this paper, we develop a semantics-regularized dictionary learning method to incorporate such advanced semantic information into the training model to improve material identification performance. A set of optimization algorithms is developed to obtain the solutions of the proposed optimization problem. Finally, we perform extensive experimental evaluations on publicly available datasets to show the effectiveness of the proposed method.