AI expert shares insights on creating robot with physical capabilities to beat humans in popular game


Artificial intelligence has been able to beat masters at games like chess and poker and Go. 

AI has also been able to beat human competitors in various video games. While impressive nonetheless, there is one major capability that these games do not require of the AI: physical skill. 

CyberRunner is an AI tasked with learning how to play the popular labyrinth maze game. 

This game requires motor skills, spatial reasoning abilities and a lot of practice for humans who play. 

WHAT IS ARTIFICIAL INTELLIGENCE (AI)? 

“I have always thought this is a really interesting game where you need fine motor skills,” Raffaello D’Andrea, a professor at ETH Zurich in Switzerland who was a lead researcher on this project, told Fox News Digital. 

“It’s a physical task … It’s not like chess, or Go, which you can play on a computer. This is the physical world, and that has always interested me,” D’Andrea said of choosing the labyrinth game for the AI to learn. 

Raffaello D’Andrea is a professor at ETH Zurich in Zurich, Switzerland. He is also a founder of Kivia Systems, which was acquired by Amazon, and is now Amazon Robotics. (Verity)

The popular game is essentially a maze in which players need to get a marble from start to finish without the marble falling into the many holes that are found throughout the board. 

This is a game that D’Andrea has not only played himself, but has gotten quite good at, he says. 

It took D’Andrea and one of his students, Thomas Bi, about a year to create CyberRunner. 

AI GIVES BIRTH TO AI: SCIENTIST SAY MACHINE INTELLIGENCE NOW CAPABLE OF REPLICATING WITHOUT HUMANS 

“AI is an art. It’s not just about, ‘OK, we’ll take the latest algorithm and apply it.’ It’s really knowing how to set up your architecture. How to set up the neural network,” D’Andrea said. 

“This is where we are right now. People are figuring it out. Figuring out the art of artificial intelligence — and this is really what we did together.”

AI sign

“AI is an art,” Raffaello D’Andrea said. “People are figuring it out.” (JOSEP LAGO/AFP via Getty Images)

Their AI project ended up being extremely proficient at the task it was given. 

It was able to complete the maze in record time with just six hours of learning. CyberRunner beat the record time set by a human player by over 6%. 

“What was really surprising to us was how well it worked,” D’Andrea said. “We did not think that in six hours, it would be able to be the world’s best at this. It was a big surprise to us.”

AI IN YOUR SHOES? THIS COMPANY AIMS TO ACCESSORIZE EVERYTHING YOU OWN WITH ARTIFICIAL INTELLIGENCE 

This project is going to be open-sourced, which D’Andrea said will allow the technology to serve as a teaching tool. 

Open-sourcing this technology provides an inexpensive avenue for everyone from high school students to top researchers to explore more about AI. 

Labyrinth game

CyberRunner (pictured here) was able to set a record time for completing the maze. “CyberRunner beat the record time set by a human player by over 6%.”  (ETH Zurich)

“We’re open sourcing the project. Anyone can duplicate what we’ve done,” D’Andrea explained. 

Many research labs conducting similar research use expensive hardware that is not widely accessible. 

CLICK HERE TO SIGN UP FOR OUR LIFESTYLE NEWSLETTER

“A lot of times it’s bespoke hardware. Hardware that can only be used in their lab. No one else can can really validate it or leverage it. I can imagine this being in every high school,” D’Andrea said. 

AI brain model

AI is beginning to accomplish more human-like tasks, and CyberRunner is an example of this. (iStock)

A successful AI like CyberRunner that can complete a complicated given task efficiently opens the door for many AI advancements in the future. It serves as proof that robots can accomplish tasks that take a lot of practice and skill to achieve. 

CLICK HERE TO GET THE FOX NEWS APP

“I really do think that this is an existence proof that shows that you can use a deep reinforcement, trained AI to learn a physical task, which is usually the purview of human beings,” D’Andrea said. 

“Machines are good at doing things fast, doing repetitive tasks,” D’Andrea said. 

“You can have a robot that drills a screw super fast, but there’s not much skill involved in doing that. It’s just raw speed. This is not a skill of raw speed. It’s really about fine motor skills. So I think we’ve shown existence proof that one, you can create AI’s that learn to do these tasks, and two, that the learning can actually happen really quickly.”

For more Lifestyle articles, visit www.foxnews.com/lifestyle.



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *