Analyzing the motions of a new robot

1
25

Over the first eight weeks of summer, I conducted research here at Union under the guidance of Associate Professor of Computer Science John Rieffel. For those eight weeks, I was paid by the school through a NASA grant to research how to make a certain kind of robot move.

I worked on a type of robot called a “tensegrity” robot, which is a portmaneau of “tensile” and “integrity.” This reflects the key fact about these robots which make them so interesting: they maintain their shape using the tensile pressure of springs, rather than a single rigid structure.

The robot I worked was named VALTR (Vibrationally Active Limbless Tensegrity Robot) and was made of six rigid bars, each held to the others by small springs. The reason VALTR and other tensegrity robots are so interesting is because they can be deformed — squished, pulled, bounced, and so on — without sustaining much damage.

When they experience forces like that, the springs holding them together stretch rather than break. As soon as the force is removed, the springs get tight again and the robot snaps back into its normal shape.

This means that tensegrities can be easily stored, since they can be compressed, and they deploy again very quickly. They can also be easily dropped from heights, such as onto the surface of another planet.

The CROCHET Lab’s VALTR tensegrity robot. Courtesy of Union College Computer Science Department
The CROCHET Lab’s VALTR tensegrity robot. Courtesy of Union College Computer Science Department

Tensegrities has one major problem in its relative immobility. Because they are so vibrationally active, it’s impossible to predict how they will react to stimuli. Moreover, you can’t put wheels, treads, or legs on them since that defeats the purpose.

Instead, Professor Rieffel and I worked on using vibrating motors to move the robot. When the robot vibrates at specific frequencies, it moves around the table, much like a phone set to vibrate. My job was to use a certain kind of computer algorithm, called Bayesian optimization, to try and find the frequencies which were the best at moving it forward.

In the end, the results were inconclusive, so I will be continuing the research this year.

1 COMMENT

Leave a Reply