Such a system will cost at least 5 times less and will allow to navigate an electric vehicle quickly and with high accuracy in space regardless of weather conditions, while its energy consumption will decrease by 18%.
Scientists from South Ural State University managed to integrate a machine vision system and automated control of electric transport into a single system, which not only increased traffic safety, but also optimized the use of energy by such a car,
With the help of the Matlab Simulink program, using a convolutional neural network, the researchers developed an effective mathematical model and algorithms for analyzing the image transmitted from light-sensitive sensors to orient an object in the surrounding space.
Based on the machine vision system, a mathematical model and algorithms for the control system of a prototype of an electric unmanned car were also created. Optimization of handling processes allowed scientists to reduce the energy consumption by transport by 18.7%. It is noted that the resulting model demonstrates high reliability and meets all the energy efficiency requirements for electric vehicles.
"We have created a prototype of an electric race car control system and conducted functional tests to check its performance in various scenarios: adaptive speed control with calculation of time before collision, decision-making to brake or avoid an obstacle," said Artem Rauer, developer, Master's student in Mechatronics and Robotics at SUSU. "Our project implements a hybrid approach for continuous and unimpeded planning of a car's trajectory, which guarantees that it will find its way. The mathematical model we created has successfully passed the standard compliance test, demonstrated its performance and has prospects for further development."
Among the advantages of the new control system is its cost, which is reduced by at least 5 times, thanks to the use of cameras, while maintaining the functionality of measuring distances since expensive laser distance meters (lidars) costing from 200 thousand to several million roubles are not used.
As compared to its analogues, the system by the Chelyabinsk developers also stands out thanks to its work on the principle of focal vision, which allows only a narrow zone around the observed object to be processed in high resolution (similar to the human directed gaze). This makes it possible to reduce the load on the computing power of the system and increase the speed of information processing.
The most accurate positioning of an object in space is achieved, in addition to GPS, by processing information from cameras. This approach allowed the researchers to minimize errors in determining the spatial position of an electric vehicle.
In the process of improving the project, the researchers are working on improving the accuracy of object recognition by increasing the training sample, as well as introducing advanced data expansion methods.
Receiving a grant for the project implementation will allow the researchers to move on to creating a beta version (preliminary version of the product) for comprehensive testing of the system on a real prototype of an unmanned car in various road scenarios and climatic conditions.
The project by the Chelyabinsk scientists has commercial potential in the segments of autonomous transport and intelligent transport systems. The implementation of the development will be of interest to manufacturers of electric vehicles and special equipment, managers of urban infrastructure projects and specialized race car series.