Engineers “bestowed” unmanned machine thinking

Now the drones will have common sense.

Scientists from the Massachusetts Institute of technology presented startup iSee, which develops unmanned vehicles based on the new approach to artificial intelligence systems. Instead of relying on traditional machine learning algorithms, programmers are inspired by cognitive Sciences that will allow them to endow unmanned transport semblance of common sense in people and the ability to quickly navigate in unexpected situations. About the project tells MIT Technology Review.

The drivers behind the wheel are based on knowledge of the characteristics of the surrounding world, when you make decisions on the road. For example, they know that the bus need more time to slow down, and that it can suddenly go from lots of pedestrians. To create an onboard computer that could respond to such a situation, using traditional methods is very difficult.

For example, last year in China’s self-driving car Tesla at the speed entered in machine for sweeping streets according to some experts, this happened due to the fact that the onboard computer was trained in Israel or Europe, where no such variety of cleaning cars.

The founders iSee Debbie Yu (Debbie Yu), Ibio Zhao (Zhao Yibiao) and Chris Baker (Chris Baker) are going to understand how a person retrieves information about the world, to create a system of artificial intelligence of a new type. Programmers have not disclosed details, but tell you what, in particular, they explore intuitive perception of the physics of the world — in a similar area of work, researchers from Google DeepMind.

An understanding of how the world works around you, allows you to predict how you will develop an unexpected situation. In addition, according to Josh Tenenbaum (Josh Tenenbaum), counsel for the company, knowledge of the laws of the physical world is closely connected with the intuitive understanding of psychology and ability to predict what goals the person — for example, reaching for a Cup of tea to take her.

A new startup combines classical methods of machine learning (particularly deep learning) with probabilistic programming. As a platform for testing the researchers used Lexus RX450h that belonged to one of the founders of iSee. As reported on the website of the company, the prototype test has already traveled thousands of kilometers and had a record low number of failures. However, the company has not yet provided any more specific statistics.

In recent years, more and more scientists are working to teach artificial intelligence systems and robots to interact effectively with objects in the world. So, Google has managed to teach a robotic hand-eye coordination in the capture of real objects, and to determine characteristics of the objects in the virtual world.

paradox