3-D Dense Reconstruction of Vision-Based Tactile Sensor With Coded Markers

Perceiving accurate 3-D object shape is an essential and challenging task for robotic manipulation, which is commonly based on vision systems. However, vision perception suffers from several limitations, especially in manipulation tasks where objects are often occluded by the robotic hand. Alternatively, tactile perception attracts a lot of attention. Due to the low resolution, the density and efficiency of existing tactile-based 3-D reconstructions are limited. In order to solve the above problems, this article describes a vision-based tactile sensor with coded markers. By combining the neighborhood structure coding method and U-net-based decoding algorithm, the sensor can reconstruct high-density 3-D object shapes efficiently. Extensive experimental results show the promising sensitivity, accuracy, stability, and robustness of our proposed sensor.