Environmental obstacle detection and localization model for cable-driven exoskeleton

Abstract

The cable-driven exoskeleton robot is an assistive device to help older people with their daily walking, so it needs to recognize and locate obstacles in its walking path and generate proper gaits. Models that use single-source data can only achieve recognition or localization separately. Its accuracy is also lower than expected. Therefore, it cannot meet the needs of exoskeletons. In this paper, a deep learning model based on multi-source is proposed for the lower limb ankle cable-driven exoskeleton. A multi-source dataset with matching RGB and depth images is also established to make the exoskeleton perceive obstacles and determine their location simultaneously. Finally, the model’s effectiveness is verified by experimentally recognizing different-sized obstacles and calculating their spatial coordinates. The model’s accuracy of recognition and localization reached 92% and 0.02m, respectively.