What is Tensegrity Robotics?

A/N: Most of the links are short youtube videos and I definitely recommend watching them. So much cool science.

I was working in lab on Tuesday and on the lab-top someone had left open an issue of Soft Robotics to a specific page. It showed the picture of a guy and the title of the article was “An Interview with NASA Principal Investigator Vytas SunSpiral: Expert Opinion on the Advantages and Limitations of Soft Robotics”, by Barry Trimmer.

Barry Trimmer, a soft robotics professor at Tufts University here in Boston and editor-in-chief of the Soft Robotics journal, likes to interview people in the field and ask them about their work, recent field developments, the future of robotics, things like that.

Vytas SunSpiral received his B.A. in Symbolic Systems and his M.S. in CompSci at Stanford. He currently works for NASA and works in an incredibly small and recently developed area of robotics, called Tensegrity Robotics. It actually has nothing to do with soft robotics, exactly–yet. The two areas may be melded together in the near future, but for the time being, both areas are currently coming into their own realms.

So, call me extremely clueless when my colleague sees me perusing the open page and asks me a question about tensegrity, as I had no idea what the word even meant at that point. From there I went home, read the article, and then printed and read about 8 more articles from the last 3 years (all listed at the end of the blog), just trying to understand the history and scope of this field.

This area of research is just so amazingly cool that I decided to write a my first official blog post about it.

Tensegrity, the word itself, is a blending of ‘tension’ and ‘integrity’. The robots in this field are not hard robots (basically the robot everyone is familiar with–hard shells, heavy duty motors at every joint, etc.), nor are they soft robots (made of silicones and shape memory alloys with no metal involved). Instead, tensegrity structures are comprised of rods, which are loaded solely in axial compression, and cables, which are loaded solely through tension. The lengths of the cables are adjusted through the use of motors at the ends of the rods, thus producing conformational changes in the overall structure of the robot.

Here is Vytas’ current project, the Super Ball Bot, to give you an idea of what these things look like.

NASA’s Super Ball Bot.

Here is another tensegrity robot, called the Tetraspine. It’s made of tetrahedron-shaped segments and each node (where a rod meets another rod) is connected to at least two and at most four other nodes by cables. And here’s a video of it moving in simulation.

Simulation of Tetraspine tensegrity bot in a platform called Bullet Physics Engine.

Not quite what you imagine when you think “robot”, huh? The ends of the rod hold motors that can release or increase tension on any given cable. This fluctuating tension is what gets the robot to move. (From there, you get into evolving control algorithms to find the most efficient way to move, but that’s a little too in-depth for this blog post and I’m probably (definitely) not well-versed enough to be able to adequately explain them. The papers I read that deal with these topics are included at the bottom.)

Another key point of tensegrity robots is that none of the rods are connected to one another. This does two things: it first removes any lever arms in the system, which could (intentionally or not) magnify external forces on the structure. It also allows for externally-applied forces to be transmitted along multiple pathways, rather than just along one, and this allows the structure to distribute forces in a way analogous to soft robotics, rather than hard robotics.

The trouble with hard robots is that they are so fragile and touchy. You may think, they’re metal! They’re machines! How can they be fragile? Well, it turns out that in order to get these insanely complicated machines to move, the control over these machines has to be so exact and finite that often times a small perturbation – a pebble in the way of a walking robot, a slight incline in the floor, a breeze just strong enough to push the robot a millimeter to the side – is enough to wreak havoc and cause a crash.

Here is my favorite DARPA robotics competition video. So, what’s the problem?

The problem is that iron-fisted control. When you want a robot to behave like a being, like a human or an animal, but you program it like a robot, you have…issues.

Screenshot of a video showing all the muscles that are utilizing during the process of standing from a sitting position.

Here’s the difference. When you decide to get up from your chair and walk across the room to grab your 5th cup of coffee that morning, do you think about each individual action? “Place feet flat on floor, equidistant to ensure balance. Activate all the muscles mentioned in this nifty youtube video. Rotate torso forward to keep center of mass (COM) centered to prevent falling backwards when muscles are activated. Push with legs muscles to standing position.” Then you get to walking part and that’s just an entire mess of actions and COM considerations. What about pouring the coffee? “Extend arm first by rotating at the shoulder, then integrate elbow extension to achieve maximum arm length. Position hand around coffee handle, curl fingers inward and use sufficient closing strength to prevent dropping the coffee. Tighten arm muscles to lift the pot….”  Yes? No? Probably no. You just think, I want to get coffee, and just…do it. 

And when you break it down like I did above, movement in animals is actually pretty amazing. Your entire body knows what to do and when to do it without you having to spend any effort at all actually thinking about it. (Theoretically, we know how animals move [paper and video], but reproducing it goes into the realm of those complicated algorithms I don’t want to talk about).  So, right now, in order to get the robot to move, you have to program and code for each and every single movement and adjustment in the bot. Until we can get a robot to think about movement the way humans think about movement–i.e., not at all–we’ll continue to have issues.

And god forbid they fall over and something breaks. Another thing these robots lack is what we call “robustness”–the ability to persist and be resistant to damage. Animals and humans can take a licking and keep on ticking. Incorporating robustness is yet another factor that needs to be improved upon in robots.

People are making great strides in robotics–check these bots out–but there’s room for a lot of improvement. A/N: There has been a lot of progress in hard robotics both before and after I wrote this post, and they can do incredible things. But they are still super complicated machines, whereas soft robots and tensegrity robots are often less complicated and easier to build. 

So, what can we do? The robots that we have aren’t great at reproducing realistic movement. We can try to make them more realistic to begin with, starting with the materials they’re made from. Are animals comprised solely of metal? No, they’re an amalgamation of rigid support structures and soft and squishy tissue. This brings us to soft robotics–the incorporation of soft materials and eradication/limitation of hard components. It’s the flip side of hard robotics. You can imagine hard robotics to be focusing solely on reproducing the skeletal systems in animals, while soft robotics is working on reproducing the muscles. It’s still in its very early stages and most of the work is focusing on simple systems, like the caterpillar, or fingers.  But soft robotics has made some improvements regarding robustness and we’re doing pretty okay with the whole “moving thing” so far. A lot of soft robotic application is focusing on gripping, or using soft actuators in place of hard motors to get something to move (my PI has a paper in review about this, so I’ll write another blog when it is published).

Reproduced from the Whitesides Research Group’s website: ‘A 9-cm tip-to-top pneu-net [soft robotic actuator] gripping an uncooked chicken egg. A string suspends the gripper and assists in lifting the egg; a tube, visible on the left side of the gripper, runs in to the central portion of the gripper to provide pressurized air for actuation.’ (Source)

But where does tensegrity robotics fit into all of this? It starts off in biology.

The human body modeled to highlight the idea of tensegrity. (Source)

Have you ever done a really good calf stretch and felt relief in your lower back? Or more
generally, had an injury in an isolated part or your body, like the foot, and then started feeling aches in your hips or shoulders? Traditionally we have thought of the body as a group of isolated muscles, supporting a skeletal system. But a new ideology is taking over, “tensegrity of the body”, in which we kind of look and operate just like the super ball (or rather, the super ball looks and operates just like us). Our bones are the rods and our muscles are the cables (arm and leg videos). When you have a system that is this interconnected, flexible and adjustable, a perturbation can be easily distributed throughout the system, passed from rod to cable to rod until it’s evenly distributed and weak enough in any one given spot that it can’t do any damage.

When you land from jumping down, if you keep your legs straight, it really hurts! When you bend your knees and allow the force of landing to propagate up your legs, it might not even hurt at all. Same idea. Because hard robots are nothing but metal and motors connected very rigidly (for that fine-tuned control they seek), there is no force distribution, it can’t handle perturbations to the system and it breaks.

With a tensegrity robot, it can move around and absorb impacts in ways that distribute the load around the entire structure, rather than through one main pathway. It is robust, it is simple, it is biologically realistic. What’s not to love?

The current plans for the Super Ball Bot are to send it to Mars as the next evolution of rovers. One of the current issues with rovers is that, well, they’re hard robots. In order to safely deliver them to the surface of the planet, you need to have a lot of padding and a big parachute. All that padding and parachute adds a lot of weight for what is essentially a one-time-only use. It’s not very efficient. But what if you have a robot that can drop from a pretty decent height–like, oh say, the atmosphere–and survive without damage? No need to pack a parachute. Instead, use the extra space in the rocket to pack 10 more of these super ball bot suckers.

The rovers are pretty limited in what areas they can explore because they’re so fragile. Can’t go up hills. Can’t go near cliffs. Can’t go over rocky terrain. These super ball bots can do all of those things without fear of being damaged. Near a cliff? Hell, let’s just roll the thing off the cliff and see what’s at the bottom. Why not? It’ll survive.

Somewhere off in the future, tensegrity robotics and soft robotics will get together and have a marvelous love child and we’ll be that much closer to producing living machines. I can’t wait to see (and maybe even be involved?) when it comes around.




P.S. – I just learned that this blog is the second result when ‘tensegrity robotics’ is googled. That is super amazing, and I know a lot of people find my site through that search. If there is anything I can do to improve or update this post, and make it worth being Second Result, I’m more than happy to hear your suggestions. Thank you! 

Reference Papers (Chronological Order):

  1. Iscen, A., Agogino, A., SunSpiral, V., and K. Tumer. Controlling Tensegrity Robots through Evolution. GECCO’13. 2013.
  2. Tietz, B.R., Carnahan, R.W., Bachmann, R.J., Quinn, R.D., and V. SunSpiral. Tetraspine: Robust Terrain handling on a Tensegrity Robot Using Central Pattern Generators. AIM. 2013.
  3. Bruce, J., Caluwaerts, K., Iscen, A., Sabelhaus, A.P., and V. SunSpiral. Design and Evolution of a Modular Tensegrity Robot Platform. ICRA. 2014.
  4. Iscen, A., Agogino, A., SunSpiral, V., and K. Tumer. Flop and Roll: Learning Robust Goal-Directed Locomotion for a Tensegrity Robot. IROS. 2014.
  5. Martinez, R.V., Glavan, A.C., Kelinger, C., Oyetibo, A.I., and G. Whitesides. Soft Actuators and Robots that are Resistant to Mechanical Damage. Advanced Functional Materials,  24: 3003-3010. 2014.
  6. Mirletz, B.T., Bhandal, P., Adams, R.D., Agogino, A.K., Quinn, R.D., and V. SunSpiral. Goal-Directed CPG-Based Control for Tensegrity Spines with many Degrees of Freedom Traversing Irregular Terrain. Soft Robotics, 2(4): 165-176. 2015.
  7. Iscen, A., Caluwaerts, K., Bruce, J., Agogino, A., SunSpiral, V., and K. Tumer. Learning Tensegrity Locomotion Using Open-Loop Control Signals and Coevolutionary Algorithms. Artificial Life, 21: 119-140. 2015.This one is more like a review paper and is pretty easy to digest. Highly recommend.
  8. Mirletz, B.T., Park, I., Quinn, R.D., and V. SunSpiral. Towards bridging the reality gap between tensegrity simulation and robotic hardware. IEEE. 2015.
  9. B. Trimmer. An Interview with NASA Principal Investigator Vytas SunSpiral: Expert Opinion on the Advantages and Limitations of Soft Robotics. Soft Robotics, 2(2): 51-58. 2015. —Definitely read this one, at least. It’s well written and explains a lot of the history and ideas involved at an easy-to-understand level. 
  10.  D. Rus and M.T. Tolley. Design, fabrication and control of soft robots. Nature, 521. 2015. –Really good review of soft robotics. 
  11. Lessard, S., Castro, D., Asper, W., Chopra, S.D., Baltaxe-Admony, L.B., Teodorescu, M., SunSpiral, V., and A. Agogino. A Bio-Inspired Tensegrity Manipulator with Multi-DOF, Structurally Compliant Joints. Arxiv. 2016.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s