Read more: Visit websiteRobots and AI might pose a threat to your job, but I doubt Cristiano Ronaldo and Lionel Messi are losing sleep over it — or so it seems from a new robot demonstration by DeepMind, Google's AI research lab.
In a new deep reinforcement learning paper spotted by TechCrunch , DeepMind researchers showcase a duo of miniature humanoid robots playing a one-on-one game of soccer.
The researchers set out to find whether deep reinforcement learning is an effective training method for teaching robots sophisticated movement skills and complex behavioral strategies at a low cost. So what better way to put this hypothesis to the test than a game of soccer.
For the uninitiated, deep reinforcement learning is a machine learning training method that lets computational agents — like AI and robots — learn by trial and error without much pre-programmed input from human engineers.
DeepMind says the robots were ultimately able to pick up skills like "rapid fall recovery, walking, turning and kicking" along with smooth transitions between movements. They were also able to develop a sense for the dynamics of the game, learning how to anticipate ball movements and block shots.
"Animals and humans are not just masters of their bodies, able to perform and combine complex movements fluently and effortlessly, but they also perceive and understand their environment and use their bodies to effect complex outcomes in the world," the paper notes.
Although the proof-of-concept bred success, DeepMind adds the next step is to pit robots against each other in teams — and eventually even play the game by solely relying on their built-in sensors in environments without any external motion capture systems.
Google has placed multiple bets on the future of smart humanoids — an increasingly competitive landscape that has yet to deliver on its promise.
No comments:
Post a Comment