An End-to-End Haptic Adjective Recognition Method with Self-Attention Mechanism

Abstract

Human’s ability to describe the tactile sensation of objects has piqued the interest of numerous researchers seeking to augment the dexterity of robots in delicate tasks. However, most existing approaches are limited by their two-stage framework, resulting in low inference efficiency and unsatisfactory performance. To address this challenge, we propose the first end-to-end framework for haptic adjective classification. Specifically, our framework leverages the Space Encoding Module to capture long-term dependencies, and the Order Encoding Module to learn order information explicitly. We conduct experiments on the public PHAC-2 Dataset and the result shows that our method achieves F1 score of 0.759, outperforming previous work in a significant way.