Like other Disney creations, Jimmy looks rather magical.
While humanoid robots can be painfully slow, Jimmy moves with lifelike speed and grace. A video posted earlier this year shows the robot waving at people, doing a little dance, drumming on a table. Just as impressive, Jimmy can safely operate near people, and by “near” we mean in contact with them. In the video, the robot plays patty-cake with a kid and even pats her cheeks—something you don’t see very often in human-robot interaction experiments.
There’s no magic, of course, just beautiful engineering. Jimmy is not powered by the bulky electric motors and gears commonly used in humanoid robots; instead, it relies on a new kind of actuator designed by Disney researchers that consists basically of tubes filled with air and water. And while the current version of the robot requires a human puppeteer to control its movements, future models could be made fully autonomous.
To find out more about Jimmy, we spoke with John P. Whitney, who led the development of the robot while at Disney Research in Pittsburgh; he’s now a professor of mechanical and industrial engineering at Northeastern University in Boston, where he’ll continue working on this technology.
VERY FLUID MOTIONS
We first wrote about Disney’s fluid actuators a couple of years ago. Whitney and colleagues from Disney Research, the Catholic University of America, and Carnegie Mellon University have since improved their design.
The actuator uses two fluid transmission lines, one with air and one with water. The air transmission (thin red tube in the picture above) works as a preloaded spring and damping system, while the water transmission (thicker black tube) actuates the joint. Each transmission is attached to cylinders containing a rolling diaphragm. The diaphragm moves back and forth, pushing a rod. A gear system (red geared pieces) converts the rod’s linear motion into rotation. The actuator weighs only 120 grams, and can deliver up to 4.5 newton meters of continuous torque with a 135 ◦ range of motion.Photo: Disney Research
The original actuator used either air or water. Now an enhanced, hybrid configuration uses both fluids to deliver more speed and torque [photo, right]. Whitney says the device has greater torque per weight (torque density) than highly geared servos or brushless motors coupled with harmonic drives. It’s also compliant and backdrivable, making it intrinsically safe—and thus ideal for human-robot interaction applications.
When people meet Jimmy for the first time, Whitney says, most feel “a strong emotional connection” with the robot. “It’s always amusing to hear people describe his motions as very fluid!”
THE MAN BEHIND THE CURTAIN
The main advantage of this kind of actuation system is that, unlike motors or servos, you don’t have to place the entire system inside your robot’s limbs, so you can make them smaller and lighter. But there are disadvantages too: Like any fluid-based system, you need to regularly check on its pressure levels. And more significant: To build an autonomous robot, you’d need a set of motors and a control system capable of replacing the human puppeteer who’s manually driving the fluid actuators [below].
Jimmy is operated by a puppeteer hiding behind a wall. To make Jimmy’s arms move, the puppeteer uses a replica of the robot as a controller. The replica is coupled to the actual robot through a series of air and water transmission lines, which transfer forces from one side to the other. Not only can the puppeteer move Jimmy’s arms but he can also haptically feel when the robot’s arms touch things or get pushed. A stereo camera on Jimmy’s head streams video to a head-mounted display worn by the puppeteer, and servos on the robot’s neck replicate the puppeteer’s head motions. Eventually, a control system and a set of motors could be used on the puppeteer’s side to make Jimmy’s movements fully autonomous.Photo: Disney Research
We’d love to see Jimmy and other robots like it in Disney parks soon. But we’re also hoping that fluid actuators could find applications beyond entertainment. (Whitney says no patents on this technology have been filed in his name, and since he’s no longer affiliated with Disney Research he can’t speak to any follow-on work or Disney’s intellectual property policies.) In particular, it would be great to see some experiments with this kind of actuation in the personal robots space, which could use some breakthroughs in manipulation.
Whitney says that’s definitely an area he’s looking into. “The haptic benefits to a human operator are equally valuable for autonomous control,” he says, “and the backdrivable and lightweight properties of the transmission are great features to have when you adopt manipulation and ambulation strategies that leverage rather than avoid contact with the environment.”
Here’s our full interview with him.
IEEE Spectrum: This approach to robot actuation is radically different from the usual motor and gear system. Where did the idea come from?
John P. Whitney: The original motivation was the same as for the MIT WAM arm and other impedance-based systems designed for human interaction: Using a lightweight high-performance transmission allows placing the drive motors in the body, instead of suffering the cascading inertia if they were placed at each joint. An ultralight arm can move much more swiftly and delicately, and we have the freedom to use larger motors with low gear ratios to minimize added friction and preserve backdrivability. But unlike a fixed-cable transmission, we can easily add more degrees of freedom to a system by simply running additional flexible hydraulic lines—fluid-systems scale incredibly well to high-DOF systems.
There is also a strong motivation to develop simpler and more economical systems, because lightweight, high-ratio, low-friction gearboxes (cycloidal, harmonic, cascading planetary) are extremely expensive. Since our drives are “off-board,” we can get away with a heavy-but-inexpensive solution.
The nonmotorized “puppet” configuration began as an easy way to test and demo the system, but it has become an interesting approach with unique applications in its own right.
Is this kind of actuation used in other systems?
Yes! Rolling diaphragms are used for lubricant-free industrial automation, optical table suspensions, and locomotive brakes—they have been around for about a century. However, nearly all uses are pneumatic, except their use in precision hydraulic pressure regulators. These diaphragms are typically rated for much lower pressures than industrial hydraulic systems, and economical manufacturing methods (fabric-reinforced compression molding of sheet rubber) limit the stroke that these actuators can achieve, so they aren’t indicated for most industrial applications.
Our contribution was using these diaphragms in a hydrostatic configuration, driven manually or with electric motors, rather than pumps and valves. In our recent ICRA paper we have figured out a hybrid air-water hydrostatic configuration that simplifies the system even more.
What was your reaction the first time you put together a prototype and it worked?
When first testing the water-filled version of the system, the perfect matching of the input and output and the ability to move at very high speeds was quite eerie—almost like looking into a live 3D mirror!
Although the original prototype transmission was much heavier, in a single-joint demo this doesn’t impact performance. We discovered the haptic qualities of the transmission only by accident. Someone suggested we try rubbing a rough surface to see if we could feel it through the transmission, and we were quite amazed that you could.
From the videos it’s clear that the system is fast and robust. Are there any drawbacks—for example, does it need maintenance?
The diaphragms have proven extremely durable. Over time they will (in theory) wear out, but the lack of rubbing, and the careful design to minimize fatigue of the reinforcing fibers, contributes to very long life. Because the diaphragms are soft and fiber reinforced, they tend to age gracefully.
Over time there is a small amount of transpiration of water through the diaphragm, so about once a month we recalibrate the amount of fluid in the transmission to recenter it, but this is a 30-second procedure. Initially filling the system with water and bleeding all the air out is challenging, but this will be addressed in the next design iteration, to make this process easy and fast.
Are you going to be able to continue working on this approach at Northeastern?
There are many exciting avenues for continuing this work, and potential high-impact applications outside of entertainment and social robotics research. One interesting area is MRI-compatible remote needle biopsy, and other MR-guided procedures [the actuators require no ferrous materials or metals]. These applications are interesting particularly because of the ability to give the clinician real-time force feedback, concurrent with 3D MR video.
I am also very excited about applications for autonomous manipulation—the ability to be swift yet delicate, and the natural sense of environmental “proprioception” enabled by an ultralight and ultralow friction robot arm will allow for manipulation strategies that embrace contact with the environment, rather than a traditional vision-only approach that must carefully avoid any unintended contacts.
The system seems ideal for teleoperation, but are there any plans to make it autonomous, maybe using motors on the operator side?
The original motivation for the “puppet” configuration was that it allowed us to separate the hardware and mechanical challenges from the perception and artificial intelligence challenges. Being able to simulate, with a human operator, the performance of future-tech level AI is incredibly powerful; it provides an existence proof of physically realizable performance.
Now, with a well-tested mechanical system, we can revisit autonomous operation—the original vision for this technology—with confidence that our hardware will not restrict what can be achieved. We are learning that many of the “analog” qualities of this system will pay dividends for autonomous “digital” operation; for example, the natural haptic properties of the system can be of equal service to an autonomous control system as they are to a human operator.