GelLight: Illumination Design, Modeling, and Optimization for Camera‐Based Tactile Sensor

Abstract

Camera‐based tactile sensors attract the attention of the robotics communities by the high‐density tactile perception, in which image quality and reconstruction accuracy are significantly determined by the illumination design. However, the influence of illumination has not yet been systematically analyzed, and most existing sensors adopt empirical design and subjective evaluation to determine the light configuration. Herein, a photometric stereo‐based modeling, optimization, and evaluation system is proposed to explore the best illumination for typical camera‐based tactile sensors. First, this article constructs a tactile benchmark dataset, simulates the contact deformation of elastomer surface, rendering the tactile imaging under various illuminations, and constructs a metrics system to evaluate the performance. Then, the relationship between reconstruct accuracy and illumination direction distribution on the benchmark is depicted, and the best illumination is optimized. The optimized sensor is fabricated and evaluated by standard metrology experiments, which exhibits high reconstruction accuracy and convincingly demonstrates the effectiveness of the proposed design and optimization approach. Furthermore, intensive experiments are conducted on diverse objects, which additionally indicate the generality and adaptability of the designed sensor. Herein, the illumination design can simplify and improve the performance of camera‐based tactile sensors.

Keywords

• camera‐based tactile sensors
• photometric stereo
• robot perception
• tactile perception