Improving the Representation and Extraction of Contact Information in Vision-Based Tactile Sensors Using Continuous Marker Pattern

Abstract

Tactile perception has been a hot topic of research in robotics. Robots sense the shape, material, distributed force, slip during contact, and use the multi-modal contact information to control grasping and manipulation. For vision-based tactile sensors, the contact representation and extraction determine the quality of the raw tactile information, and therefore serve a significant role in the robot perception system. This article highlights for the first time the importance of raw representation and extraction in visuotactile perception, and proposes a new multicolor CMP method for enhancing the performance of vision-based tactile sensors. Based on the principle of continuous marker pattern (CMP), the multicolor CMP method is optimized in the pattern and algorithm design. Regarding information representation, we present a new type of marker pattern based on RGB triangles and a preferred layout. In terms of information extraction, we propose a series of extraction strategies with the adaptive growing algorithm (AGA) and the spin-search algorithm (SSA) as the cores. The experiments reveal that the multicolor CMP method achieves improved precision and reliability compared to the former CMP method.