Scientific research and technology often go hand-in-hand, resulting in ground-breaking developments that revolutionize industries. An enlightening example of this is the recent accomplishment by a team of researchers at Delft University of Technology. They've developed an autonomous drone that leverages neuromorphic image processing and control, inspired by the functioning of animal brains.
Conventional deep neural networks require a copious amount of data and energy. This poses logistical problems for small drones, as they can't efficiently support large hardware or an extensive battery system. It's here that the recently developed neuromorphic processors come into play. These processors have exhibited extraordinary performance during flight, processing data up to 64 times faster and consuming three times less energy compared to typical GPUs.
The promise of this technology is that future drones could become as small, agile, and intelligent as insects or birds. This could lead to drones being able to handle a broader range of tasks while overcoming the limitations of space and energy.
Drawing insights from how animal brains function, specifically their asynchronous information processing and sparse processing, researchers have developed spiking neural networks. These networks replicate how biological neurons communicate via electrical pulses or spikes, and they minimize energy consumption as sending such spikes require energy.
Compared to standard deep neural networks, these spiking neural networks are faster and more energy-efficient. That's not all - when used with neuromorphic sensors like cameras, these networks can function in both dark and bright environments, recognize motion quickly, and are more energy-efficient. All of these features combined can majorly boost autonomous robots, particularly agile robots like drones.
Researchers marked a milestone by developing a drone that uses neuromorphic vision and control for autonomous flight. Accounting for the drone's pose and thrust, they developed a spiking neural network that processes signals from a neuromorphic camera and gives out control commands. Incorporated into a drone along with Intel's neuromorphic research chip, Loihi, the drone can perceive and control its motion in all directions.
An autonomous drone equipped with neuromorphic vision and control can fly at different speeds and under varying light conditions. It even shows resilience to flickering lights, which typically cause the neuromorphic camera pixels to send irrelevant signals to the network.
The utilization of Neuromorphic AI goes far beyond increasing the speed and energy efficiency of drones. Its potential is vast, and it could render smaller autonomous robots more capable. Such tiny drones can be used for a variety of applications, from managing inventory in warehouses to monitoring crops in greenhouses.
These small drones are not only cost-effective and safe but can also navigate in narrow environments. Additionally, the deployment of these drones in swarms can significantly enhance the coverage of an area. But the realization of these applications depends on further shrinkage of the neuromorphic hardware and evolving capabilities for complex tasks like navigation.
Disclaimer: The above article was written with the assistance of an AI. The original sources can be found on ScienceDaily.
