Remoto Pilot

The Bright box company which developed connected car of the solution Nissan Smart Car for the Middle East, KIA Remoto and Infiniti in Russia and more than 90 mobile applications for 450 dealer centers announced in November, 2016 development of the system of pilotless driving of the car, working using studying at the generated personnel from computer games and real frames of shooting from streets of a neuronet of computer vision.

Remoto Pilot is the system of pilotless driving of the car allowing in real time the car to detect and go round obstacles, to go on in advance planned trajectory, guaranteeing safe and reliable driving on a band. A key technology of the experimental solution of Bright Box company is use of stereosight in combination with advanced algorithms of computer vision on the basis of neural networks. Use of the global navigation satellite system (GNSS) and high definition (HD card) in a system gives the chance of completely independent operation of the car.

See Also:

• Auto pilot (unmanned vehicle)
• The system of computer vision Cognitive Technologies for pilotless cars
• Pilotless KAMAZ is expected until the end of 2017
• AgroBot Pilotless tractor

Enhanced capabilities of technologies of computer vision, such as convolution neural networks (CNN) in a combination to stereoscopic sight, can raise possibilities of independent driving of cars considerably. The technology used in the solution of Bright Box company gives the chance of the safe and reliable movement in real time, detection and a detour of different obstacles (cars, pedestrians).

Use of stereosight (pair of video cameras installed on the car) allows to calculate distance to different objects in sight of the camera. Thus, there is possible assessment of a road situation the on-board computer of the car also in real time.

Training of neural network

Training of neural network assumes use of previously collected and processed sets of video samples of road situations, the called training data sets. For preparation of such sets records from the dash cameras mounted on cars which go on roads are kept. Besides, in the solution such modern technology of training of neural network as 3D computer graphics from computer games is used. Such games as GTA-V where the most part of a gameplay is a driving on city streets, with a large number of realistic views of streets including from within inside of the car, become a valuable source of high-quality images which can be used for creation of training data sets.

The data sets and the synthesized data sets written on real driving from computer games represent very effective set of the educational samples used for training of neural networks for the analysis of real road situations and providing safe and reliable autonomous driving on city streets. Big advantage of such approach is the flexibility inherent in neural network algorithms. Thanks to the above described approach, it turns out to analyze much bigger quantity of road situations.

Neural networks can be also trained for measurement of distance from a stereo pair of cameras to different objects using a set of training data which include the images written from the stereocameras set on real cars when driving on real roads.

One of widespread types of the sensors used in autonomous cars – it a lidar (the laser scanner which measures distance to surrounding objects), as a rule, is installed on the car on a roof. Use of a stereocamera alternative to a lidar. Stereocameras, as well as lidars, are used for measurement of distances to objects, and can be also to be applied to creation of deep cards of planning of a trajectory of the car. Placing emphasis on use of modern methods of computer vision, Bright Box to aim to develop technologies which will allow to exclude need of use of lidars.

"We have a successful experience in development of connected car of solutions for remote control functions of the car, and more than 250,000 connected cars go with our technology. Today we work with a broad base of clients of car makers worldwide (Europe, the Middle East, Asia). The company already conducts negotiations by car makers and partners in delivery of a new solution and hopes for increase in their quantity in the future", - the technical director of Bright Box, Dimchenko Alexander said.

The solution Bright Box is suitable for retrofitting of already released cars.