Computer Vision-Based Pose Estimation of Tensegrity Robots Using Fiducial Markers

Abstract

Tensegrity paradigm can allow creation of lightweight, robust, efficient and configurable robots. Modeling, design, control and sensing of tensegrity robots are challenging. Effective control of tensegrity robots requires accurate state information in real-time. In this work, we address the problem of six-degree-of-freedom pose estimation of tensegrity robots. Due to the unorthodox structure of the tensegrities and their wide range-of-motion, a computer vision-based technique is employed. To the best of our knowledge, this is the first work which uses computer vision to determine the internal state of a robot. Specifically, a hemispherical camera module was attached to the base of the robot pointing toward the top part of the robot. This camera was used to track the fiducial markers printed on a triangle-shaped plate on the upper part of the robot. Position and orientation of this plate in 3D-space was obtained by the use of ArUco markers, a fiducial marker system originally devised for augmented reality. The efficacy of our fiducial marker-based pose estimation framework was shown with extensive real-world experiments with a six-bar tensegrity prism robot. Specifically, the average position and orientation estimation errors for a 1.5 m tensegrity were 2.3 cm and 7.5 degrees, respectively.