Deep-Learning-Assisted Underwater 3D Tactile Tensegrity

Peng Xu, Jiaxi Zheng, Jianhua Liu, Xiangyu Liu, Xinyu Wang, Siyuan Wang, Tangzhen Guan, Xianping Fu, Minyi Xu, Guangming Xie, and Zhong Lin Wang; RESEARCH.

Abstract

The growth of underwater robotic applications in ocean exploration and research has created an urgent need for effective tactile sensing. Here, we propose an underwater 3-dimensional tactile tensegrity (U3DTT) based on soft self-powered triboelectric nanogenerators and deep-learning-assisted data analytics. This device can measure and distinguish the magnitude, location, and orientation of perturbations in real time from both flow field and interaction with obstacles and provide collision protection for underwater vehicles operation. It is enabled by the structure that mimics terrestrial animals’ musculoskeletal systems composed of both stiff bones and stretchable muscles. Moreover, when successfully integrated with underwater vehicles, the U3DTT shows advantages of multiple degrees of freedom in its shape modes, an ultrahigh sensitivity, and fast response times with a low cost and conformability. The real-time 3-dimensional pose of the U3DTT has been predicted with an average root-mean-square error of 0.76 in a water pool, indicating that this developed U3DTT is a promising technology in vehicles with tactile feedback.