Robot first responders could adapt to damage in hostile settings

Scientists have created robots that can keep moving even after sustaining what would normally be debilitating damage, a resiliency that could help them survive as first responders in situations too dangerous for humans.

"We are constantly amazed at how much damage they can handle," Jeff Clune, a University of Wyoming assistant professor of computer science, tells CBC's Quirks & Quarks in an upcoming interview. Clune co-authored a paper published in Nature showcasing the research.

• Listen to an interview with Clune on CBC's Quirks & Quarks this Saturday
• Read the paper in Nature

Clune was part of a team of researchers who developed a six-legged machine that looks like a spider, and a mechanical arm with multiple joints.

Normally, robots are programmed to do a specific task, and if their body changes, that program no longer functions, he says. Clune and his team wanted to create robots that could continue their tasks until they could be brought in for repairs.

Raising a robot toddler

Clune and the other scientists gave their robots a simulated childhood. The scientists created a computer simulation of the robot and ran it for two weeks. Much like human toddlers play to figure out how their bodies work and learn how to walk, the robots played to figure out the different ways they could move effectively.

In the simulated world, the robots tried out more than 40 million different behaviours, he says, and retained the successful ones.

The researchers challenged the robots to come up with as many different movements as possible. The six-legged robot was asked to walk with all its legs touching the ground, with its front right leg never touching the ground and numerous other combinations.

The robot even managed to surprise the team with its creativity. It was asked to walk without any of its feet touching the ground, without any expectation this was possible. But the robot simply flipped onto its back and crawled using its elbows, says Clune.

Adapt within 2 minutes

The researchers then tested the robot's versatility with different injuries. Since the robots are expensive, the researchers often simply disabled a motor to simulate damage.

Clune says at that one point they tried sawing off a leg from the six-legged robot.

Robots are actually learning in real time right in front of you.— Jeff Clune, study co-author

The robot realizes it's damaged, but it has no knowledge of where its injury stems from. It retrieves a behaviour from its simulated childhood to test out under the new conditions. Often, its first attempt isn't successful, so the robot recalls another behaviour learned in its childhood.

"It will increasingly try a few more to find better and better behaviours," says Clune, "until, eventually, it looks like it's limping away like a wounded animal."

• AI could destroy humans, Stephen Hawking fears: Should you worry?
• 15 humanoid robots who will do your work for you

It takes less than two minutes for the robots to adapt to the damage they've sustained, he says.

The researchers also tested the robots in slightly more challenging conditions, including on a slope and on slippery, freshly waxed floors. The robots also adapted well to those conditions.

"For the first time, we think, you know robots are actually learning in real time right in front of you. They're not taking hours and hours, and conducting thousands of trials," he says. "In a handful of tests, in about, you know, a minute or two, they're up and walking."

Future first responders

The researchers believe the algorithm has the potential to work on any robot, he says. They're looking forward to applying it to more elaborate robots and putting them in more complex environments.

Eventually, these robots could become first responders to crises in a hostile environment.

"Robots are going to provide tremendous benefits to society once they can do tasks that are too dangerous for humans."

The robots could help fight forest fires, search for survivors after an earthquake or go into cities after a nuclear disaster, like the tsunami-crippled Fukushima nuclear power plant in Japan in 2011.

In those situations, it's likely the robots would sustain damage. But Clune says they could continue to soldier on and save lives despite their injuries.

"In all those cases, we should be sending robots."

Seemingly indestructible robots are unlikely to take over the world Terminator style, says Clune. The robots can simply be turned off.

A robot that can shrug off any obstacle thrown its way would have tremendous value, he says. It's "an absolutely great thing if that robot is finding survivors in an earthquake zone or putting out a fire. You want it to keep going."

However, he admits that the technology they've created could be further developed to create something for nefarious purposes.

"The reality is that any tool or any technology can be used for great good or great evil, and that depends on the people that wield it."