Evolution of Locomotion on a Physical Tensegrity Robot

Due to their high strength-to-weight ratio, robustness and de-formability, tensegrity structures are an appealing platform for the emerging field of soft robotics, with applications ranging from search-and-rescue to field-deployable structures. Unfortunately, these properties also make tensegrities challenging to control through conventional means. In this paper we describe a novel means of vibration-based tensegrity actuation which allows for the manual control of a physical tensegrity robot in the plane as well as state-machine based target following. We demonstrate the evolution of effective gaits using only physical evaluations of the robot, and further demonstrate how a combination of the state-machine with the hill climber allows for the hands-off automation of the evolutionary process. We conclude with a description of how this can lead to a bootstrapping effect, with the potential to significantly accelerate and automate the physical evolution of our tensegrity robot.