Drones can already navigate independently on a free field, but autonomous flying is not yet possible for complex applications and environments or in places where there is no sufficient GPS signal. A research team around Jan Steinbrener from the University of Klagenfurt comes in: with the help of artificial intelligence (AI), learning drones, such as living things, gradually find their way in their environment.
A big challenge is that every autonomous navigation system must always know exactly where it is, said Steinbrener, head of the project funded by the Science Fund FWF. A GPS signal, on the other hand, has an accuracy of two to three meters no means that is sufficient for a safe flight near Huizen or critical infrastructure. In addition, drones can quickly become unstable, so you often have to estimate your position for a calm flight.
AI -filters incorrect data
In the case of camera images alone, a calculation of the position would be needed a hundred times per second for a stable flight. The small computer on the drone, on the other hand, only manages to evaluate the photos about ten times per second. That is why they are combined with gear sensors. “These are simple sensors that make small, easy -process data available -because the data is usually also bad in terms of quality, this means that the accuracy of the calculation decreases rapidly over time,” says Steinbrener.
The AI algorithm plays in the game: the system is intended to reduce and act as a filter for interference factors. On the other hand, the algorithms can also help to analyze dark or washed camera images.
Successful “Live Training”
The training of the algorithms is not carried out, as is often the case with a powerful computer and previously recorded data, but “live” on the drones. “A toddler only gradually learned how it can use its own body to find his way around the world,” the researcher said. Likewise, the algorithms must learn how to help with navigation. To protect the fragile devices, they were protected with a cable for crashed during the first flight tests.
At the moment the drone succeeds in driving autonomously without much exercise and the results of the gear data are better than with “classic” approaches, Steinbrener continues. In this way have performed the “live training” of the algorithms, is a great success. As the next step, the researchers are planning to stabilize more complicated flight maneuvers using AI.
The advantage of this is flexibility in terms of the Drohen model used. “It would be extreme time to collect a lot of data from any other drone, to train on a large computer and then implement – with our procedure the algorithm could be trained on the respective model for a few minutes and would then be ready for use,” said Steinbrener.
From a regulatory point of view, still open questions
According to the researcher, there are many areas of application: from the inspection of difficult to reach, critical infrastructure such as electricity pylons or bridges to the search for missing people in extensive areas for continuous monitoring of large areas, for example in agriculture.
For special applications, especially indoors, the system could be available in the near future. This is more complicated outdoors: from the point of view of regulatory, many questions are still open. From a technical point of view you are on the right track. “I hope that in a few years we will be ready in outdoor areas – also with regard to the regulations,” summarized Steinbrener.
Source: Krone
I am an experienced and passionate journalist with a strong track record in news website reporting. I specialize in technology coverage, breaking stories on the latest developments and trends from around the world. Working for Today Times Live has given me the opportunity to write thought-provoking pieces that have caught the attention of many readers.
