A team of researchers at Delft University of Technology has developed a drone that flies autonomously using neuromorphic image processing and control based on the workings of animal brains. Animal brains use less data and energy compared to current deep neural networks running on GPUs. Neuromorphic processors are therefore very suitable for small drones because they don’t need heavy and large hardware and batteries. The results are extraordinary, with the drone’s deep neural network processing data up to 64 times faster and consuming three times less energy than when running on a GPU. Further developments of this technology may enable drones to become as small, agile, and smart as flying insects or birds.
Artificial intelligence holds great potential to provide autonomous robots with the intelligence needed for real-world applications. However, current AI relies on deep neural networks that require substantial computing power, especially for small robots like flying drones. Animal brains process information asynchronously and communicate mostly via electrical pulses called spikes, leading to sparse processing and energy efficiency. Inspired by these properties, scientists and tech companies are developing new neuromorphic processors to run spiking neural networks that promise to be much faster and more energy efficient, making them ideal for small autonomous robots like drones.
The researchers at Delft University of Technology developed a drone that uses neuromorphic vision and control for autonomous flight, deploying a spiking neural network on Intel’s Loihi neuromorphic research chip on board of the drone. The network allows the drone to perceive and control its own motion in all directions. The team faced challenges in training the network but designed a network consisting of two modules that learn motion perception and control commands. Through artificial evolution in simulation, the spiking neural networks got increasingly good at control and were able to fly in any direction at different speeds, functioning well on the real robot.
The neuromorphic vision and control of the drone enable it to fly at different speeds under varying light conditions and even with flickering lights. Measurements confirm the potential of neuromorphic AI, with the network running significantly faster and more efficiently on the neuromorphic chip compared to an embedded GPU. This improved efficiency allows for deployment on much smaller autonomous robots, showcasing the potential of neuromorphic AI for tiny robots. Future applications may include monitoring crops in greenhouses, tracking stock in warehouses, and exploration and gas source localization in narrow environments with swarms of tiny, safe, and affordable drones.
The development of neuromorphic AI is seen as an absolute enabler for tiny autonomous robots, allowing for more intelligent and efficient operations. The advantages of tiny drones include their safety, ability to navigate in narrow environments, affordability for deployment in swarms, and quicker area coverage. The current work at Delft University of Technology is a significant step in this direction, but further scaling down of neuromorphic hardware and expanding capabilities towards more complex tasks such as navigation will be essential for realizing applications in monitoring crops, tracking stock, exploration, and gas source localization. Overall, neuromorphic AI has the potential to revolutionize the capabilities of small autonomous robots and enable a wide range of applications in various industries.