Disney Research is teaching this skinny robot how to perform a variety of somersaulting stunts
This is a guest post. The views expressed here are solely those of the author and do not represent positions of IEEE Spectrum or the IEEE.
Olympic gymnast Simone Biles has a signature move that is named after her because she is the only woman on earth capable of performing it. The move starts as a layout double flip, but more than halfway through suddenly develops a twist that rotates her body through an extra 180 degrees to land face first. The only visible source of this sudden change in rotation is a small motion of one hand as her arm goes from straight to bent. It’s a beautiful example of how the seemingly simple physics of ballistic motion, completely governed by a relatively simple conservation of angular momentum, can produce amazing and unexpected results.
Disney Research has a rich history of pushing the boundaries of innovation at The Walt Disney Company. That said, we found that combination of simple physics and surprising results compelling—perhaps it would be possible to build a relatively simple robot that could accomplish remarkable things.
Additionally, we saw two potential benefits of building robots that could perform acrobatic stunts while aloft. First, a robotic performer can answer questions about how a performance is accomplished that are more difficult to answer with a human performer. The robot has a small, known set of actuators and sensors, an easily measured and adjustable weight distribution (a robot can put on five pounds even faster than I can), and no free will to do anything other than what’s asked of it. Together, those traits make it easier to pull out single variables and test to what degree each of them contributes to a successful move.
Second, the force, speed, and precision required to execute acrobatic maneuvers pushes the limits of robot capability in a way that has the potential to be relevant to the broader field. Acrobatic moves involve high centrifugal forces, short decision times, and challenging launch and landing dynamics—all of these can drive us towards a deeper understanding of actuator, sensors, and control strategies.
A year ago, the Disney Research team produced a very simple, brick-shaped robot that could move some weights internally to affect its spin rate and steer itself through a narrow slot when dropped from the ceiling. In our upcoming ICRA paper, we present the next step along the path towards an acrobatic robot: A three-link, pneumatically actuated device that’s capable of performing a variety of simple acrobatic behaviors in a predictable, controllable fashion.
“Stickman” is a z-shaped robot about seven feet in length and weighing about 40 pounds. The frame is made of 80-20 and is capable of folding up tightly or extending almost into a straight line using two pneumatic actuators. In doing so the robot changes its moment of inertia and hence its spin rate, much like an ice skater pulling her arms in to accelerate her rotation. In our experiments, we attach one end to a long steel rod, hoist it up to the ceiling, and drop it.
The robot swings forward and lets go of the rod, sending it flying out over some foam mats. After a flip or two, the robot lands on the mats in a target configuration—usually stretched out parallel to the ground to distribute the force of impact and minimize the chance of damage to itself.
This launch method is a nice way of injecting energy smoothly into the system: Gravity gives us a nice acceleration and the pendulum gently arcs our velocity into the right direction. The pneumatic actuators helped achieve the high power density required to tuck quickly under high centripetal loads, and the 80-20 aluminum construction made for easy adjustment and repair.
To perform intelligently, the robot uses two sets of sensors: an inertial measurement unit, or IMU, and an array of three laser rangefinders. The IMU uses gyroscopes and accelerometers to estimate the orientation of the robot relative to the ground and to measure the robot’s angular velocity. When the IMU thinks the rangefinders are pointed at the ground, they take range measurements. Each of the three is mounted at a slightly different angle so they “see” the ground at slightly different times, helping us to estimate not only the robot’s height but also its velocity.
Once we have a good estimate of height and position, Sir Isaac Newton gives us a good sense of when we’re going to hit the ground. To land in a good orientation, we just have to know how fast we’re spinning and how much we can change that rate by tucking or untucking. In the paper we show an ability to hit a target orientation within a few tens of degrees; in the future we hope to do much better.
We also found that the dynamics of the swing were surprisingly rich, and that by exploiting those physics in addition to our mid-flight control we were capable of performing single backflips, double backflips, and flying through the air with almost no rotation at all.
Right now our two actuators don’t give us a full two degrees of rotational control because they are mounted in-plane. If we simply added one more link whose axis of rotation was at 90 degrees to our current joints, it’s possible to start influencing twist in new and potentially interesting ways.
We’re still a long way from a robot capable of performing “the Biles,” but we are already encouraged by how the elegantly simple physics of motion have elevated our simple robotic stick into something capable of a legible acrobatic performance. We hope that we can continue to build our scientific understanding of the rich and fascinating opportunities created by tossing a robot across the room.
Morgan Pope is a postdoctoral associate at Disney Research in Glendale, Calif. He joined Disney after graduating from Stanford, where he focused on building small robots that could jump, perch, and climb.
Source: IEEE Spectrum