Research Status and Development Trends of Depth Perception Technologies for Transparent Objects in Robotic Applications

Accurate environmental perception is the foundation for robots to perform operational tasks. Currently, robots typically utilize depth cameras (such as infrared cameras, lidars or binocular vision systems) to obtain depth information about the environment. However, transparent objects possess non-lambertian properties. The structured light of depth cameras is prone to specular reflection and refraction on their surfaces, resulting in some light being unable to return or returning incorrect signals (corresponding to non-real surfaces), thereby causing the loss and error of depth perception and ultimately leading to the failure of robot operation. To address the aforementioned challenges, this paper reviews the current research status and development trends of robot depth perception technology for transparent objects from four aspects:hardware, datasets, methods, and evaluation indicators. Research shows that the multimodal information obtained by robot sensors can accurately predict the depth of transparent areas, improve the perception and operation accuracy of robots, which is of great significance for the reliable deployment of robots in complex scenarios such as hospitals, factories, and chemical laboratories, and is conducive to promoting the development and application of intelligent robots with bodies.