Artificial intelligence has great potential to give autonomous robots the intelligence needed for practical applications. However, current AI relies on deep neural networks that require significant computing power. A research team from TU Delft recently obtained a drone developed They can fly independently through so-called neural image processing and control, which relies on the work of animal brains and consumes less energy.
Neural processors are therefore very suitable for small drones because they do not require heavy and large hardware and batteries. It turns out that animal brains use less power than current deep neural networks running on graphics processing units (GPUs). Test results are promising. In flight, the drone’s deep neural network based on neuromorphic technology processes data up to 64 times faster and consumes three times less power than using a GPU. Further developments of this technology could lead to a leap forward in drones becoming as small, agile and intelligent as flying insects or birds. The results were recently published in Scientific robotics.
Inspiration from animal brains
Animal brains process information in a very different way than neural networks running on GPUs. Biological neurons process information asynchronously and communicate mainly via electrical impulses, which are called “spikes” in English. Because sending such waves requires energy, the brain reduces these waves, resulting in scattered processing.
Inspired by the properties of animal brains, new neural processors are being developed by scientists and technology companies. “While spiking digital neurons only need to add whole numbers, scalar neurons need to multiply and add decimal numbers. This makes spiking neural networks faster and more energy efficient. After all, people find calculating 5 + 8 much easier than calculating 6.25 x 3.45 + 4.05 x 3.45?,” says doctoral student and one of the authors of the article, Jesse Hagenaars.
Neural cameras
Power efficiency is further increased when neuromorphic processors are used with neural sensors, such as neural cameras. These cameras do not record images over a specific period of time. Instead, each pixel only emits a signal when it becomes brighter or darker. The advantages of these cameras are that they can detect motion much faster, are more power efficient, and work well in both dark and bright environments. Furthermore, signals from neural cameras can be fed directly to spiking neural networks running on neural processors. Together, these technologies could enable the development of autonomous robots.
Over generations of artificial evolution, spiking neural networks have become increasingly better at controlling. They were finally able to simulate a drone in any direction and at different speeds. “We trained the two modules and developed a way to integrate them together. We were pleased to see the combined network working well immediately on a real robot,” say the researchers. Thanks to neuroimaging and control, the drone is able to fly at different speeds under different lighting conditions, from Dark to bright.
Future applications
According to researchers, neural AI makes all autonomous robots smarter. “But it is an absolute game-changer for small autonomous robots. At the Faculty of Aerospace Engineering at Delft University of Technology, we are working on small autonomous drones that can be used for applications ranging from monitoring crops in greenhouses to tracking supplies in warehouses. The advantages of small drones are Pilot in that they are very safe and can navigate tight spaces like rows of tomato plants, plus they can be very cheap, so they can be deployed in swarms, where we need to rescue people or Professor of bio-inspired drones, Guido De Croon, explains how Quickly locate gas leaks.
Drone in healthcare
The use of drones is increasing. They are increasingly used not only by individuals and companies, but also in the healthcare field. Last summer, the Albert Schweitzer Hospital conducted a test drone flight between the hospital and the healthcare facility’s warehouse. University Medical Center Groningen (UMCG) has also investigated whether medical drones could be used to bring care closer to the patient. UMCG has joined forces with UMCG for this purpose Medical drone service.
