If you’ve been to the RoboGames, you’ve seen everything from flame-throwing battlebots to androids that play soccer. But robo-athletes are more than just performers. They’re a path to the future.
Researchers at the University of Electro-Communications in Tokyo and the Okinawa Institute of Science and Technology have built a small humanoid robot that plays baseball — or something like it. The bot can hold a fan-like bat and take swings at flying plastic balls, and though it may miss at first, it can learn with each new pitch and adjust its swing accordingly. Eventually, it will make contact.
The robot, you see, is also equipped with an artificial brain. Based on an Nvida graphics processor, or GPU, kinda like the one that renders images on your desktop or laptop, this brain mimics the function of about 100,000 neurons, and using a software platform developed by Nvidia, the scientists have programmed these neurons for the task at hand, as they discussed in a recent paper published in the journal Neural Networks.
Yes, it’s fun. But through this baseball-playing robot, the scientists also hope to better understand how brains can be recreated with software and hardware — and bring us closer to a world where robots can handle more important tasks on our behalf.
When a ball is pitched to the robot, an accelerometer at the back of a batting cage records information about the flight of the ball, including its speed, and this data is relayed back to a machine that holds the GPU-powered brain. The brain then crunches this data so that it can determine exactly when the robot should swing. If the scientists change the pitch speed, the robot will relearn the task all over again.
This is not the first time researchers have modeled a cerebellum to control robots. A team of scientists in Europe, for instance, have used an artificial cerebellum to control a robotic limb. But according to Tadashi Yamazaki, one of the scientists who worked on the project, the baseball-playing robot is the second largest model of its kind and it runs in realtime, meaning its much faster than other systems. That means the GPU brain is better suited to controlling external hardware, he says.
Just as he did with a previous artificial brain model, Yamazaki plans to release the source code for the system. The whole idea is to push this area of research forward. “Working code helps other scientists to learn how to implement an articial brain in computers,” he says.