With electronic games to train unmanned systems?
The ford of the research and innovation center in palo alto, California, on a big screen, a robot in the shape of a character running faster than a four-wheel drive car.It sometimes unexpectedly ran in front of the car and even to push the car parked on opposite sides-to hold the road.The speed of the human role running and turning accidentally read faster than reality.
The system we are trying to defeat our fellow Tory Smith explains.
Tory Smith's team is using the game development tool to build virtual environment to create a series of action sequences can enter machine learning system.In this way, the researchers hope to self-driving cars software can better deal with real road situations you might encounter.(by the way, this day is not far off: ford CEO Mark Fields has just announced that ford has approved, ready to test prototype of driverless cars in California in 2016.)
Researcher Ashley Micks said, the use of virtual environment, will be able to rapidly promote learning speed.Need to 10 days before the mission, now only 20 minutes, she said.
Ford researchers developed simulation software not only can run in the big games system, can also run on the phone.Smith said: "we can put the mobile version of random distribution to the pedestrians in the street."He said it would be very useful, because the outside can come up with many research team of strange situation and challenge driverless cars.In the game environment, the software can not only learn how to avoid pedestrians, can also learn all kinds of knowledge, for
example, there is no lane markings on the ground under the
condition of how to determine where the driveway.