Watch AI-powered robots play football

Google DeepMind can now train small, ready-made robots to be on the football field. In a new article published today in Science Robotics, researchers describe their recent efforts to adapt a subset of machine learning known as deep reinforcement learning (deep RL) to teach bipedal bots a simplified version of the sport. The team notes that while similar experiments in the past created extremely agile quadrupedal robots (see: Boston Dynamics Spot), much less work has been conducted on bipedal, humanoid machines. But new footage of the bots dribbling, defending and shooting goals shows just how good a deep reinforcement learning coach could be for humanoid machines.

Although it is ultimately intended for enormous tasks such as climate prediction and materials scienceGoogle DeepMind can also absolutely destroy human competitors in games like to play chessgo, and even Starcraft II. But all those strategic maneuvers don't require complex physical movements and coordination. So while DeepMind can study simulated football movementsit has not been able to translate to a physical playing field, but that is quickly changing.

To create the miniature Messis, engineers first developed and trained two deep RL skills in computer simulations: the ability to get up from the ground and how to score goals against an untrained opponent. From there, they virtually trained their system to play an entire one-on-one football match by combining these skills and then randomly pairing them with partially trained copies of themselves.

[Related: Google DeepMind’s AI forecasting is outperforming the ‘gold standard’ model.]

“So in the second phase, the agent learned to combine previously learned skills, refine them into the full football task, and predict and anticipate the opponent's behavior,” researchers wrote in their introduction. They later noted that “During play, the agents switched fluidly between all of these behaviors.”

AI photo

Thanks to the deep RL framework, DeepMind agents quickly learned how to improve existing skills, including how to kick and shoot the football, block shots, and even defend their own goal against an attacking opponent by using his body as a shield .

During a series of one-on-one competitions in which robots took advantage of the deep RL training, the two mechanical athletes walked, turned, kicked and stood faster than if engineers simply provided them with a baseline of skills. These weren't minuscule improvements, either: compared to a non-customizable script baseline, the robots ran 181 percent faster, turned 302 percent faster, kicked 34 percent faster, and took 63 percent less time to get up after a fall. Furthermore, the deep RL-trained robots also exhibited new emergent behaviors, such as turning and spinning on their feet. Such actions would be extremely challenging to prescribe otherwise.

Screenshots of robots playing football
Credit: Google DeepMind

There's still some work to do before DeepMind-powered robots reach RoboCup. For these initial tests, researchers relied entirely on simulation-based deep RL training before transferring that information to physical robots. In the future, engineers want to combine both virtual and real-time reinforcement training for their bots. They also hope to scale up their robots, but that will require much more experimentation and refinement.

The team believes that using similar deep RL approaches for football, as well as many other tasks, could further improve bipedal robots' movements and real-time adaptation capabilities. Still, it's unlikely you'll have to worry yet about DeepMind's humanoid robots on big football fields – or in the job market. At the same time, given their continued improvements, it's probably not a bad idea to get ready to blow the whistle.

Related Posts

The mindset, technology and tools holding back e-commerce

Powered by TeleSign Ecommerce fraud is on the rise — here’s how to stop it. Read this on-demand VB Spotlight to gain insight into the most common types of fraud.…

NASA's VIPER rover won't go to the moon after all

In November 2019, NASA reported announced plans to send a new rover to the moon. However, after nearly 5 years and multiple delays, it appears that Volatiles Investigating Polar Exploration…

Leave a Reply

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

You Missed

Pelosi has told House Democrats that Biden may soon be convinced to withdraw from the race

  • July 19, 2024
Pelosi has told House Democrats that Biden may soon be convinced to withdraw from the race

18 Silk and Great Value plant-based milk alternatives recalled in Canada due to listeria deaths and illnesses

  • July 19, 2024
18 Silk and Great Value plant-based milk alternatives recalled in Canada due to listeria deaths and illnesses

Trump Bitcoin Conference Fundraising Tickets for Nashville Soiree Hit $844,600 Max

  • July 19, 2024
Trump Bitcoin Conference Fundraising Tickets for Nashville Soiree Hit $844,600 Max

Solar parks with rainwater management reduce runoff and erosion, research shows

  • July 19, 2024
Solar parks with rainwater management reduce runoff and erosion, research shows

Food aroma research could help explain why meals in space taste bad

  • July 19, 2024
Food aroma research could help explain why meals in space taste bad

The mindset, technology and tools holding back e-commerce

  • July 19, 2024
The mindset, technology and tools holding back e-commerce

Writer Shalom Auslander catalogs his lifelong battle with self-contempt in ‘Feh’ : NPR

  • July 19, 2024
Writer Shalom Auslander catalogs his lifelong battle with self-contempt in ‘Feh’ : NPR

Canadian Jacob Shaffleburg, Colombian Richard Rios and Copa America stars set for transfers

  • July 19, 2024

Trump speaks at RNC amid Biden election questions

  • July 19, 2024
Trump speaks at RNC amid Biden election questions

Stocks Making the Biggest After-hours Moves: ISRG, NFLX, PLUG

  • July 19, 2024
Stocks Making the Biggest After-hours Moves: ISRG, NFLX, PLUG

The Key to a Purposeful Life in a Distracted World

  • July 19, 2024
The Key to a Purposeful Life in a Distracted World