When developing a new simulator, it is important to constantly verify with real hardware implementations that the resulting simulations are a reasonable reflection of reality, and not just pretty movies. We learned this early on when our first tensegrity robot simulations turned out to be violating basic laws of physics by harnessing “free-energy” generated by the unrealistic cable models built into the Bullet Physics Engine. We then spent significant time developing new and realistic elastic cable models which actually followed the laws of physics and didn’t introduce new energy into the system. In a prior paper we reported on motion capture experiments which validated that our NASA Tensegrity Robotics Toolkit matched the behavior of our six strut ReCTeR robot to within 1.3% error on position through dynamic motions.
The following video shows recent experiments to verify the behavior of our tensegrity “spine” simulations. As you will see in the following video, the basic behaviors of the simulation match well to the hardware prototype that we developed. Given that hardware is expensive to build, we made a 3 segment prototype which shows close agreement to our simulated 3-segment models, and thus we feel confident that the behavior of our larger simulated spines are realistic. The second video below shows some of those larger spine simulations which are controlled via neuroscience inspired “Central Pattern Generator” control networks.
Our full sized tensegrity spine simulations which shows their reactive adaptation to different terrains.