Towards bridging the reality gap between tensegrity simulation and robotic hardware

摘要

Using a new hardware implementation of our designs for tunably compliant spine-like tensegrity robots, we show that the NASA Tensegrity Robotics Toolkit can effectively generate and predict desirable locomotion strategies for these many degree of freedom systems. Tensegrity, which provides structural integrity through a tension network, shows promise as a design strategy for more compliant robots capable of interaction with rugged environments, such as a tensegrity interplanetary probe prototype surviving multi-story drops. Due to the complexity of tensegrity structures, modeling through physics simulation and machine learning improves our ability to design and evaluate new structures and their controllers in a dynamic environment. The kinematics of our simulator, the open source NASA Tensegrity Robotics Toolkit, have been previously validated within 1.3% error on position through motion capture of the six strut robot ReCTeR. This paper provides additional validation of the dynamics through the direct comparison of the simulator to forces experienced by the latest version of the Tetraspine robot. These results give us confidence in our strategy of using tensegrity to impart future robotic systems with properties similar to biological systems such as increased flexibility, power, and mobility in extreme terrains.