Neural calculation: The AI system will land the drone with an accuracy of one centimeter

A youth engineering team from Russia has developed an algorithm based on artificial intelligence that allows autonomous landing of unmanned aerial vehicles on a prepared site with an accuracy of up to a centimeter. The system independently determines the position and orientation of the drone in space, corrects the descent trajectory and ensures a safe landing, and the operator can only confirm the launch of the procedure. Experts believe that the technology meets the current demand for expanding the capabilities of UAVs, but a comprehensive assessment of its characteristics is needed.

Where the new development will come in handy

The first ultra-precise algorithm for autonomous drone landing in the Russian Federation was presented by the Technoshamany youth team, the press service of University 2035, which supported the project, reported. The developers have designed the control unit and created an interface for preflight configuration with the ability to automatically reboot the module.

The system is based on computer vision algorithms.: it recognizes special markers on the landing pad with an accuracy of at least 90%, while the error in determining orientation does not exceed two degrees. The OpenCV library is used to process the video stream. The frames are converted to grayscale and transmitted to the detector, the type of which is specified in the configuration file.

Currently, two types of visual markers are supported: ArUco markers and AprilTag, which are widely used in robotics and computer vision tasks.

— After detecting the marker, the system calculates its coordinates relative to the drone and the distance to it, which can reach several tens of meters. The received data is transmitted via the MAVLink protocol, which allows you to accurately control the descent of the device," explained team captain Yaroslav Kharitonov.

According to the developers, they managed to achieve the accuracy of the coordinates of the centimeter level, while there are no comparable commercial solutions on the market. The previously existing international project Vision Landing, also based on the use of markers, has been frozen.

Flying on a VR plane: An AI simulator will teach UAVs to operate without human help

How virtual reality helps neural networks make the right decisions in difficult situations

The Raspberry Pi 4 Model B is used as a computing module, which processes coordinate data. The system also includes an ArduCam camera, a flight controller, a lidar that performs the function of an altimeter, a radio receiver and a drone of its own assembly based on the Surpass engine.

— We intend to add recognition of infrared tags, as well as letters and geometric shapes on helicopter pads using neural networks. In addition, it is planned to introduce additional image processing algorithms to compensate for sunlight and increase the stability of the system in difficult lighting conditions," added Yaroslav Kharitonov.

The algorithm was presented during the training within the framework of the federal program "Cadres for UAS", operated by the university. The educational part was supervised by the Aeromobility Directorate of the Moscow Aviation Institute.

Expanding the capabilities of drones

Industry experts generally positively assess the development, noting its high application potential, but point out a number of technological and organizational limitations that need to be overcome.

Dmitry Rybakov, an Aeronet NTI expert and a member of the Energynet NTI Competence Center for the use of UAS technologies in the energy sector, called the project promising, emphasizing that such solutions were originally conceived for implementation within the framework of specialized competence centers.

— Pay attention to the experience you have already gained. Back in 2019-2020, as part of the work of Tambov State University named after G.R. Derzhavin with the participation of industrial partners, it turned out that the drone landing on visual tags is significantly influenced by external factors, in particular wind gusts. This does not allow us to fully realize the accuracy that the optical system theoretically provides, so such limitations must be taken into account at an early stage, before scaling the solution," he explained.

The expert added that the technology has a wide range of potential applications: from logistics and automated drone ports to marine operations, industrial inspection and work in the absence of GPS, for example, in forests or electronic warfare zones. In the future, such systems may form the basis of urban air mobility and air taxi standards.

Aerial hunting: "Hornets" will protect flamethrower systems from drone attacks

The new interceptor UAVs have been tested and will soon enter the army.

Nikita Danilov, CEO and co-founder of Fly Drone, also believes that the development will be in demand. According to him, it can be used in carrying out almost any civilian mission in an urban environment, as well as in areas with difficult terrain.

— It can also interest the military. However, for the full—fledged implementation of a promising project, it is important that it does not "boil in its own juice", but finds a technological partner from among the manufacturers and operators of UAVs capable of putting it into practice, the expert noted.

Varvara Gorodilova, Deputy Head of the Technological Sports Development Service at the Federal Sports Reserve Training Center, noted that the development meets an urgent demand for expanding the capabilities of unmanned systems. At the same time, she stressed the need for a comprehensive assessment: it is not only about technical reliability, but also about economic efficiency, durability and prospects for scaling production.