Image: Thomas Libby, Evan Chang-Siu, and Pauline Jennings. Courtesy of PolyPEDAL Lab & CiBER/UC Berkeley
This agile little robot with a lizard-inspired tail is by far one of the coolest things we saw at last year's IROS, the IEEE International Conference on Intelligent Robots and Systems. You better believe we wanted to share it with you back then, but the UC Berkeley researchers asked us to stay quiet since they were working on a paper for Nature, which is a pretty big deal. The embargo lifted just this second, and we're finally able to bring you the full story.
Terrestrial robots are getting to be pros at moving rapidly over, up, down, and around all kinds of surfaces, even rough ones full of obstacles. Their weakness, however, is that they generally require surface contact in order to be effective. Introduce an airborne phase (jumping, bouncing, falling, being thrown, etc.) and you get into all kinds of trouble.
Animals generally don't have issues like these. Take velociraptors, those wickedly agile biped carnivores made famous by the "Jurassic Park" movies: As I'm sure you've noticed, when velociraptors jump, they don't tip over backwards nor do they fall flat on their faces like mobile robots tend to when they get airborne at high speed. You can say the same thing about lizards if you don't happen to have any velociraptors in your neighborhood. The secret, it turns out, is use of the tail as a mechanism for generating torque to control orientation in all three dimensions, with animals dynamically adjusting the pitch, roll, and yaw of their bodies through tail motion.
Apparently the "Jurassic Park" animators got physics right in one of the movie's scenes: When a velociraptor (or perhaps more accurately a Utahraptor) jumps on a T. rex, it bends its tail upward to ensure an ideal landing posture. And by "ideal landing posture," I mean sharp pointy bits first. We don't have the clip but we found the still above.
Researchers at UC Berkeley (including Thomas Libby, Talia Y. Moore, Evan Chang-Siu, Deborah Li, Daniel J. Cohen, Ardian Jusufi, and Robert J. Full) thought that this active tail control was a pretty neat and potentially useful trick, so they performed some detailed studies of agama lizards to figure out how they did it. Using a high-speed camera, they tracked the lizards as they were, uh, persuaded to jump from a horizontal surface to a vertical one. By altering the friction of the horizontal surface, the lizards could be made to take off pitching either forwards or backwards, and close attention was paid to what they did with their tails to counteract these pitching motions to allow themselves to land on the vertical surface at the proper orientation:
Pretty cool, right? To pitch forward, the lizard simply swings its tail upwards, and conservation of angular momentum does the rest, tilting the animal's trunk "nose-up." As the researchers put it, "torque applied through the tail yields an instantaneous, predictable counter-torque on the body." Instantaneous and predictable are two very sexy words when it comes to the wild world of dynamic mobile robots, so the obvious next step was to create a mobile robot with an active tail. Meet Tailbot:
Image: Evan Ackerman/IEEE Spectrum
Tailbot (this picture shows the prototype we saw at IROS back in September) is a mobile robot of approximately the same size as a lizard. It started off as a remote-controlled car from RadioShack, weighing in at 160 grams with a length of just under 20 centimeters. The tail, of course, was not included in the RadioShack kit: It's an extra 10 centimeters long and made of carbon fiber, with a brass weight at the end of about 10 percent of the mass of the entire robot. We should note that the tail is of course modular, and both length and mass can be easily adjusted to explore different performance envelopes. The tail has its own servo, and can be driven over a range of 255 degrees in one plane up over the back of the robot (or down below it).
Control of the tail is fairly straightforward. The robot has an on-board accelerometer and gyroscope which are combined to return an estimation of the angle of the body, and then the tail is simply driven in the correct direction until the body ends up at whatever orientation you want. It's remarkably quick: After being dropped, the robot can right itself (and maintain a horizontal orientation with respect to the ground) in just one single body length as it falls. The only limitation to the system is if the bot is tumbling to such an extent (or oriented in such a way) that the tail becomes "saturated" -- in other words, it can't move far enough to exert the amount of torque required to stabilize the body. However, if the tail has a surface to push off of (like a wall or the ground), it can provide just about as much torque as you could want.
The following video shows a variety of experiments. First, you'll see some footage of the lizard jumping. Watch the tail. Then, you'll see how active tail movements alter the orientation of the robot as it goes off a jump. Lastly, there will be a couple other experiments (including a drop test and driving over obstacles) that show more examples of how useful a tail can be in real-world environments:
The UC Berkeley team next plans to endow Tailbot's tail with a yaw axis, which should also allow them to control roll through actuated tail movement. It's not at all difficult to imagine where this technique might come in handy even in its relatively simple present implementation: Think about robots like Stickybot or RiSE, both of whom depend on tails and are expensive fall risks. Adding a tail actuator and some logic could at least make sure that if (when) they do fall, they land on their feet. Also, we've got systems like the iRobot FirstLook and Recon Robotics' Recon Scout that are designed to not just drive fast over rough terrain, but also to actually be thrown places. Both of these robots have self-righting systems, but nothing as versatile as an active tail.
Image: Thomas Libby, Evan Chang-Siu and Pauline Jennings. Courtesy of PolyPEDAL Lab & CiBER/UC Berkeley
As the final piece to this puzzle, the researchers took everything that they'd learned about active tail stabilization from watching lizards jump and building Tailbot, and applied it to a morphometric model of an actual velociraptor, and found that the dinosaur "might have been capable of aerial acrobatics beyond even those displayed by present-day arboreal lizards," after which it would likely have landed on your chest and ate you before you even know it. Hear that, Spielberg? We're expecting some amazing stuff in "Jurassic Park IV."*
Oh, and for the record, we were assured that absolutely no lizards were harmed throughout the course of this research, and now animals and robots are living in harmony.
The Nature paper, titled "Tail-Assisted Pitch Control in Lizards, Robots, and Dinosaurs," is available here, and if you want to check out the research presented at IROS (there's lots more detail on the robotics side), look for "A Lizard-Inspired Active Tail Enables Rapid Maneuvers and Dynamic Stabilization in a Terrestrial Robot," by Evan Chang-Siu, Thomas Libby, Masayoshi Tomizuka, and Robert J. Full, available on IEEE Xplore.
* Please do not make a "Jurassic Park IV."
UPDATE January 9, 2012, 8:37 p.m.: Read our follow-up on Tailbot's additional tricks.