Drones Can Now Navigate Autonomously by Learning from Bicycles and Cars
Current commercial drones use GPS to navigate above building roofs and high altitude terrains. However, GPS guided drones do not have the capability to safely navigate autonomously at low attitudes in dense, unstructured streets with pedestrians, cyclists, and cars crisscrossing in their way. But that is probably going to change real soon.
Researchers at the Center for Competence in Research Robotics, Switzerland and the University of Zurich have developed DroNet algorithm. This algorithm allows drones to safely navigate on their own through streets in cities and in indoor areas. Researchers trained the drones using deep-learning algorithm on traffic using examples from cars and bicyclists.
The DroNet algorithm uses cameras on drones to enable the drone to identify and respond promptly to dangerous situations. This allows the drone to change angles and directions which keeps the drone navigating as it evades obstacles. Researchers have managed to demonstrate that newly developed drones can use the DronNet to “learn” to navigate through city streets as well as completely new environments that they were never trained to navigate before.
About the DroNet Intelligence Algorithm
The DroNet algorithm is an immensely powerful artificial intelligence algorithm that can interpret scenes around it and react appropriately. The algorithm effectively replaces the sophisticated sensors that were used in earlier models. Researchers say that the algorithm has deep learning capabilities, what they call “Deep Neural Network”. Just like kids, the algorithm has the ability to solve complex tasks by simply relying on a set of training examples or models that demonstrate how certain things are done and how to act in certain situations.
Bicycles and Cars Are the Drones Teachers
Talking of training examples’, the next time you hop on your bicycle or drive around town to pick up some groceries you might just be training a drone! When it comes to training gadgets on deep learning, one of the biggest challenges is collecting thousands of ‘training examples’. Thankfully, for this project, researchers used data they collected from bicycles and cars driving on city streets.
DroNet-equipped drones are able to automatically learn new skills by imitating bicycles and cars. Using these sophisticated algorithms, they can learn to follow safety rules like how to follow the correct lane and how to stop when they come across obstacles like other vehicles or pedestrians. Using the DroNet, drones were also able to learn how to navigate autonomously in completely new environments such as parking lots and building corridors where they had never been trained to navigate before.
The Future of Autonomous Drones
Many drone industry companies have taken note of this clever advancement and are now looking into potential applications even as the technology is still being fine-tuned. Some of the potential applications include parcel delivery, monitoring and surveillance, rescue operations in congested urban environments and more. However, while all that is not far from reality, researchers note that there is still some research and development work that needs to go into the technology to overcome a number of technological and logistical challenges before releasing it for commercial use.
|