In a major milestone for autonomous robotics, researchers at Delft University of Technology in the Netherlands have successfully developed and flown the first drone powered by full neuromorphic artificial intelligence for vision and control. The breakthrough combines bio-inspired neuromorphic processors and cameras to achieve autonomous flight with unprecedented energy efficiency and speed compared to conventional AI running on graphics processors.

The team's neuromorphic drone represents a significant leap towards making tiny, insect-like robots a reality by mimicking how animal brains process visuals and control motion in a highly optimized manner. While today's AI relies on deep neural networks that are computationally expensive, biological brains accomplish similar feats asynchronously using sparse spiking signals that are much lighter in terms of data and energy needs.

"Animal brains process information very differently than the neural networks running on GPUs," explained Jesse Hagenaars, a PhD student and author on the study published in Science Robotics. "Biological neurons communicate via electrical pulses called spikes, minimizing spiking to achieve sparse, highly efficient processing."

It's this biological inspiration of spiking neural networks that enabled the Delft researchers to develop AI algorithms capable of running on ultra-low-power neuromorphic chips like Intel's Loihi processor. Combined with a neuromorphic event camera that sends signals only when individual pixels change, rather than capturing entire frames, the full system can operate with a fraction of the compute and energy footprint of standard machine vision approaches.

"Our measurements confirm the enormous potential of neuromorphic AI compared to traditional methods," said Stein Stroobants, another PhD student involved in the work. "The network runs on Loihi up to 64 times faster while consuming just 7 milliwatts - three watts less than on an embedded GPU doing the same tasks."

In flight tests, their insect-inspired drone used Loihi's spiking neural networks to seamlessly perceive its own motion based on the event camera inputs and control thrust and positioning entirely with neuromorphic AI. It can even adapt to challenging conditions like flickering or low lights that would confuse conventional cameras.

The applications of such ultra-efficient, micro-scale autonomous robots span everything from greenhouse crop monitoring and warehouse inventory to search and rescue scenarios where flexibility and energy longevity are paramount. In short, neuromorphic AI may finally make sci-fi visions of intelligent insect-sized drones a practical reality.

"This is an absolute enabler for truly tiny autonomous robots that can be safely deployed in confined spaces or as low-cost swarms," said Guido de Croon, Professor of Bio-inspired Drones at Delft's Aerospace Engineering department. "However, realizing such applications will depend on further miniaturizing neuromorphic hardware and expanding capabilities for more complex behaviors like navigation."

While still early stage, the Delft team's breakthrough exemplifies the immense potential of neuromorphic engineering to bring major gains in autonomy, intelligence and efficiency by taking direct inspiration from the natural computational marvels found in animal brains. In the rapidly evolving field of autonomous robots, biomimicry could prove to be among the most powerful design strategies.

As researchers like de Croon continue pushing the boundaries of what's possible with neuromorphic AI for drones and other robots, society may finally begin to see intelligent, hyper-efficient machines take flight at the small scales and low-power operations once thought unattainable. The era of brain-inspired robotics has arrived.