Commercial drones use GPS to navigate, and running on top of buildings or in high altitudes is not a problem, but when drones have to shuttle between buildings at low altitudes, or suddenly break into dense unstructured city streets, cars When cycling through bicycles or pedestrians, it does not have the ability to respond quickly to unexpected events.
Researchers at the University of Zurich (Universitat Zurich; UZH) and the Center for Robotics Research at the National Center for Scientific Research (NCCR) developed the DroNet algorithm and designed a deep neural network to provide steering angles for images captured by each drone. In order to keep the drone's navigation while avoiding obstacles, and the probability of collision, let the drone discriminate the danger and react in real time.
UZH research team professor Davide Scaramuzza said that DroNet allows drones to identify static and dynamic obstacles and slow down the speed to avoid impact. It is worth mentioning that the UZH team developed the drone does not rely on advanced sensors, but uses ordinary cameras, just like on every mobile phone, but the team developed very powerful artificial intelligence algorithms to explain the drones. Observe the scene and respond accordingly.
One of the most difficult challenges of deep learning is the need to collect thousands of training cases. In order to get enough cases to train the DroNet algorithm, the research team collects data from dynamic cars and bicycles in the urban environment. Through training, the drone has automatically learned to obey the traffic rules, such as "How to travel along the lane without filling in the pair. "To the lane", "How to stop when obstacles such as pedestrians, buildings or other vehicles are blocked in front of you".
More interestingly, the research team found that their drones not only learned to shuttle between urban streets, but also learned to fly indoors, such as parking lots in the parking lot or office, and the research team never trained the algorithm to do so. There are still many technical problems to overcome in this technology. Despite this, the study also provides potential for urban street monitoring, parcel delivery, and disaster relief operations.
New Products
SHENZHEN CHONDEKUAI TECHNOLOGY CO.LTD , https://www.szfourinone.com