How autonomous cars are learning to make its way?

Autonomous vehicles can follow the general rules of the road, recognizing traffic signs and road markings, marking pedestrian crossings and other well-known features of adjusting traffic. But what to do outside of the well-marked roads, izezzhennoy length and breadth? On many roads outside cities poisterlas paint, overgrown with ivy and trees, signs, have unusual corners that are not marked on maps.

What to do following the rules of autonomous vehicle, when the rules are not clear or no? What should his passengers when they discover that their car can not take them wherever they go?

Warning hidden

Most of the problems in the development of advanced technologies include treatment of rare or unusual situations or events that require performance beyond the normal capabilities of the system. It definitely works, and in the case of autonomous vehicles. Some examples may include road navigation through the reconstruction area, meeting horse or buggy, or meeting with graffiti, reminiscent brake. Outside the road features include absolutely all manifestations of the natural world, such as trees blocked the road, floods and large pools - or even animals, blocking the way.

In the University of advanced automotive systems Center of Mississippi researchers took on the task of learning respond to the circumstances of algorithms that almost never occur, which are difficult to predict and difficult to recreate. They tried to put autonomous cars in the most complex scenario: drove the car in the area, which he had never seen before and did not know, without any reliable infrastructure such as traffic paints and road signs in an unknown environment, where you can find cactus and white with equal probability bear. In the process, they have combined the technology of the virtual and real worlds. They created the advanced simulation of realistic scenes outdoors, through which trained artificial intelligence algorithms to read the stream from the camera and classify what he saw: the trees, the sky, the open way possible obstacles. They then transferred these algorithms on a specially designed test four-wheel drive vehicle and sends it to a dedicated test site, where it is then checked the work of algorithms that collect data.

Let's start with the virtual

Engineers have developed a simulator capable of creating a wide range of realistic outdoor scenes, through which he could move transport. The system generates a variety of landscapes with different climates, forests and deserts, it demonstrates how plants, shrubs and trees grow over time. It can also simulate weather changes, sun and moon, as well as the exact position of 9000 stars.

In addition, the system simulates the reading of sensors commonly used in the autonomous vehicles, such as cameras and lidar. These virtual sensors gather data, which are then fed to the neural network as a valuable data for the study.

construct a test track

Simulations are only as good how well they reflect the real world. Mississippi State University has acquired 50 acres of land, on which scientists are developing a test track for self-SUVs. The site is great - it found the slopes at an angle of 60 degrees and a lot of different plants. Engineers have identified some of the natural features of the land, which they expect to be particularly difficult to control an autonomous vehicle, and reproduce them exactly on the simulator. This allowed them to directly compare the simulation results with real attempts to navigate the real world. In the end, they will create a similar real and virtual couple of other types of landscapes to improve the car's features.

The collection of additional data

Also, the test vehicles was created - Halo Project - with an electric motor and sensors to computers that can navigate through a variety of off-road environments. Halo Project car is equipped with additional sensors for collecting detailed data on his real environment; they help to build a virtual environment to run new tests.

The two lidar sensor, for example, under the cross angles are fixed on the front of the vehicle so that their beams are scanned approaching ground. Together they can provide information on how rough or smooth surfaces, as well as to read the data on the grass and other plants and objects on the road.

In general, scientists study gave some interesting results. For example, they showed promising hints that machine learning algorithms that train in simulated environments can be useful in the real world. As is the case with most of the research on autonomous vehicles, there is still a long way to go. Perhaps together they make self-managed vehicles are not only more functional on modern roads, but also the more popular and common method of transportation. Also waiting for? Tell us in our chatting in a telegram.