Disney Research Makes Dynamic Robots Less Wiggly, More Lifelike

A new computational method allows robotic characters to perform fast motions without excessive vibrations

2 min read
Vibration-Minimizing Motion Retargeting for Robotic Characters
Disney researchers have demonstrated a powerful new method to transfer motions created using traditional animation software to robotic characters while reducing mechanical oscillations.
Image: Disney Research

It’s an unfortunate fact that the laws of physics make it difficult to design and program robots that can move both smoothly and dynamically. The problem is that when a robotic part moves dynamically, all of the stuff that it’s attached to ends up bending and flexing. The bending and flexing may be just a little bit, if your robot is really bulky and stiff, but if you have a lightweight, compliant robot that’s designed to work around humans, the bending and flexing can be so much that it ultimately disrupts whatever motion the robot was trying to make in the first place.

Disney Research, which has an understandable interest in developing lightweight and dynamic robot characters, has presented a paper at SIGGRAPH 2019 demonstrating an effective vibration damping method for robots that would otherwise be very, very wiggly.

This fascinating video shows a sort of worst-case scenario for wiggly robots—very low stiffness structures making dynamic motions with high mass limbs. 

What you’re seeing here is not some kind of dynamic vibration damping system. In other words, if you give any of these robots a shove, they’re going to bounce all over the place. Rather, the specific motions that the robots are making (which are designed by animators) have been optimized to suppress vibrations by a computational tool described in Disney’s paper. You can’t see this happen in real time, but the tool is using a model of the robot to predict how it will vibrate, and then instruct the motors to make the very slight (but very exact) additional motions necessary to cancel out those vibrations while still making the robot move the way the animator wants it to.

You can’t see this happen in real time, but Disney’s computational tool is using a model of the robot to predict how it will vibrate, and then instruct the motors to make the very slight additional motions necessary to cancel out those vibrations while still making the robot move the way the animator wants it to

This technique does require simulation and computation in advance, and its effectiveness depends in large part on how good your model of the robot is. It gets harder and harder to do efficiently as your robot gets more complicated, and as all of the parts that can flex increases—essentially, each point of flexure introduces another degree of freedom into the mix, and since one part flexing can cause another part to flex, it quickly turns into a huge mess. Part of how the researchers tackle this is to prioritize damping out large amplitude vibrations that are the most visible. And it works very well, even on relatively complex robotic characters:

In a final demonstration, we retarget a boxing animation to the same 13-DOF full-body character, replacing the two hands with boxing gloves on both our simulation model and our physical system. Unlike the drumming sequence, the boxing motion contains faster motions with abrupt stops. The naïve retargeting causes excessive vibrations, especially when the character dodges and moves his upper body backwards and forwards. With the same objective and optimization parameters as for our Drummer, our optimized motor controls lead to deviations smaller than 1.5 cm (compared to 9 cm before optimization) while preserving the input animation without noticeable visual differences.

“Vibration-Minimizing Motion Retargeting for Robotic Characters,” by Shayan Hoshyari, Hongyi Xu, Espen Knoop, Stelian Coros, and Moritz Bächer from Disney Research, was presented at SIGGRAPH 2019 in Los Angeles.

[ Disney Research ]

The Conversation (0)

How the U.S. Army Is Turning Robots Into Team Players

Engineers battle the limits of deep learning for battlefield bots

11 min read
Robot with threads near a fallen branch

RoMan, the Army Research Laboratory's robotic manipulator, considers the best way to grasp and move a tree branch at the Adelphi Laboratory Center, in Maryland.

Evan Ackerman

“I should probably not be standing this close," I think to myself, as the robot slowly approaches a large tree branch on the floor in front of me. It's not the size of the branch that makes me nervous—it's that the robot is operating autonomously, and that while I know what it's supposed to do, I'm not entirely sure what it will do. If everything works the way the roboticists at the U.S. Army Research Laboratory (ARL) in Adelphi, Md., expect, the robot will identify the branch, grasp it, and drag it out of the way. These folks know what they're doing, but I've spent enough time around robots that I take a small step backwards anyway.

This article is part of our special report on AI, “The Great AI Reckoning.”

The robot, named RoMan, for Robotic Manipulator, is about the size of a large lawn mower, with a tracked base that helps it handle most kinds of terrain. At the front, it has a squat torso equipped with cameras and depth sensors, as well as a pair of arms that were harvested from a prototype disaster-response robot originally developed at NASA's Jet Propulsion Laboratory for a DARPA robotics competition. RoMan's job today is roadway clearing, a multistep task that ARL wants the robot to complete as autonomously as possible. Instead of instructing the robot to grasp specific objects in specific ways and move them to specific places, the operators tell RoMan to "go clear a path." It's then up to the robot to make all the decisions necessary to achieve that objective.

Keep Reading ↓ Show less