In 1940, Washington’s Tacoma Narrows Bridge, nicknamed Galloping Gertie, collapsed due to aeroelastic flutter. As wind blew above and below the bridge, aerodynamic lift was created. But, the lift wasn’t evenly distributed across the bridge, and that caused harmonic vibrations which eventually caused the structure to fail. The wind itself provided the initial energy, but it was the powerful resonance that actually destroyed the bridge. When applied in a controlled manner, that resonance can also be used for work, which is how these tensegrity robots are able to move.
Tensegrity constructs use tensile material between rigid elements to create opposing forces that result in stable structures. Using those principles, a team of researchers at ResiBots were able to create very resilient robots built from just six carbon struts and 24 springs. The design is robust enough to withstand crushing forces and significant drops, and the tensegrity robots are even able to “walk” without any traditional wheels or legs.
Using just vibratory motors, like those used to make your smart phone vibrate, the robots can learn to move across flat surfaces. As with Galloping Gertie, the vibrations resonate through the structure to amplify their strength. But, in this case, that vibration is precisely controlled to yield the desired effect: locomotion. With a trial-and-error machine learning algorithm, the tensegrity robots can learn to control their vibratory motors in order to traverse a short distance. The entire learning process can take as few as 30 trials, and it demonstrates just how powerful machine learning can be, particularly for unusual scenarios that would be difficult for humans to program explicitly.