HyQ Quadruped Robot Learns to Avoid Stumbles, Visits London

Last year, we wrote about HyQ, a quadruped robot designed for rough terrain missions. Created by a team at the Istituto Italiano di Tecnologia (IIT), HyQ was learning to walk and trot, and was also able to jump and even kick things. The robot uses hydraulic actuators, which allow it to move quickly and nimbly, with an eerie animal-like quality. Now HyQ has learned another important skill in life: how not to fall on its face when it stumbles on an obstacle.

Falling is a major problem for legged robots. Unlike animals and (most) humans, robots don't handle falls very well. Their stiff metal bodies can't absorb shocks, and a crash often means broken parts and costly repairs. So robots designed to operate in real-world conditions need to learn how to avoid falls at all costs (a prime example of that is another quadruped, the famous BigDog, and let's not forget its bigger brother LS3, both from Boston Dynamics). Cameras and sensors like LIDAR help detect and avoid obstacles, but in some situations a robot can't rely on vision—for example, in thick vegetation or if smoke is present.

To overcome this obstacle (quite literally), HyQ is learning to reflexively react when its legs hit an object on the ground. This reaction has to be very fast, especially when the robot is trotting (HyQ can reach 2 meters per second), or else it will lose its balance and collapse. The IIT researchers developed and implemented a "reflex algorithm" that allows HyQ to step over high obstacles without prior knowledge of the terrain.

"Step reflex is an important building block for a robust locomotion in challenging environments," says Claudio Semini, head of IIT's Dynamic Legged Systems Lab, where HyQ was created. He explains that the reflex they implemented in their robot is so fast that "you can't really see it unless it's shown in slow motion."

In a paper presented at the Climbing and Walking Robot (CLAWAR) conference last week, IIT researcher Michele Focchi and colleagues describe their approach and a series of tests they conducted with HyQ. The idea is straightforward: HyQ measures external forces on its feet, and whenever it detects a force acting against the motion of a foot, the step reflex is triggered.

But in practice, developing the step reflex algorithm is a bit trickier. When the foot is starting to swing forward and hits something, the robot has enough time to lift that leg and step over the obstacle. But when the foot has a collision when it's already moving down, it's too late to stop and alter its trajectory—attempting to do so could result in a fall. Instead, the robot absorbs the impact, and when it's able to perform another step, it then lifts its foot higher to overcome the obstruction. That's not a problem for HyQ because its hydraulic legs are compliant and can absorb the energy of the collision. Currently, the robot can step over obstacles up to 11 centimeters high (or 14 percent of leg length). 

The Italian researchers are now working on HyQ's vision system and on more dynamic gaits like bounding and galloping. In the next several months, they hope to test the robot in a wooded area or other rough terrain environments. In fact, one of these locations might be the Swiss Alps. The IIT group recently sold a copy of the HyQ robot to ETH Zurich's Agile and Dexterous Robotics Lab, led by Prof. Jonas Buchli. The copy has a blue torso and is called HyQ Blue [pictured below, right].

HyQ and HyQblue

And in the mean time, the original HyQ is going to London. For the first time the IIT researchers will show their quadruped to the public, as part of the Living Machines conference, which takes place at London's Natural History Museum. HyQ will be on display tomorrow, Thursday, August 1st.

[ HyQ Project ]

Advertisement

Automaton

IEEE Spectrum's award-winning robotics blog, featuring news, articles, and videos on robots, humanoids, automation, artificial intelligence, and more.
Contact us:  e.guizzo@ieee.org

Editor
Erico Guizzo
New York, N.Y.
Senior Writer
Evan Ackerman
Berkeley, Calif.
 
Contributor
Jason Falconer
Canada
Contributor
Angelica Lim
Tokyo, Japan
 

Newsletter Sign Up

Sign up for the Automaton newsletter and get biweekly updates about robotics, automation, and AI, all delivered directly to your inbox.

Advertisement