The developers of the Perm Polytechnic Institute are creating a control system for a robot that will change idle street lamps. Artificial intelligence will improve the quality of urban lighting services and avoid injury to people: now these works are carried out at altitude in adverse weather conditions. Machine learning algorithms and neural networks will help the robot navigate in space.
Scientists spoke about the development at the International Conference on Applied Innovation in IT. “Now the street lights are being changed at a height of 8 to 15 meters. In this case, work is often carried out in the presence of precipitation, wind or low temperatures. Due to the replacement of the luminaire, the transport network becomes more congested.
Specialists cannot carry out a complete diagnosis of malfunctions, as they work in an uncomfortable position and are limited in time and tools. One robotic complex will make it possible to replace several brigades that serve the lighting,”says Pavel Slivnitsin, one of the developers, a master's student of the electrical engineering faculty of the Perm Polytechnic.
To "teach" the robotic complex to navigate the city and interact with objects in the real world, scientists have developed an algorithm for it. The artificial intelligence recognizes the flashlight in the image taken by the robot's video camera and guides the robot arm to replace the lamp. Algorithms and neural networks allow the robot to solve these problems in conditions of different visibility, on different backgrounds and in a situation where the object is obscured by something.
The robocomplex is delivered to the lantern by a car, which can also be controlled by artificial intelligence. The luminaire must have a universal connector for the robot to do the job. The mount will allow you to quickly remove and install it, and replacing the lamp will become the same type of operation. The robot will perform it automatically.
In modern conditions, when intelligent devices "go" from industrial workshops to the street, it is important for them to be able to independently collect and process information, say the developers. The location of objects in the real world can change, so robots must not only work according to instructions, but also be able to act in emergency situations.
“Industrial robots in production are guided in space using predefined maps or radio tags. But in open space, outdoors, these technologies become expensive. It is impossible to tag all objects and keep the environment maps up to date. Therefore, we propose using neural networks and machine learning methods to make the robot recognize complex objects in real time. From a photo or video, he classifies the luminaires and on the basis of this makes decisions,”explains the developer.
Scientists have trained artificial intelligence on a separate server. The Permian developers trained the neural network using 250 photographs of lanterns in "familiar" angles for the robot. In the future, the selection can be replenished with new photographs that the robotic complex collects. The developers report that the device works efficiently even on computing resources with low power.
As a reminder, the first in Russia online master's degree program in robotics will start at the Perm Polytechnic Institute in 2020. Students will be able to learn how to work with service robots from anywhere in the world.