If engineers are ever going to build autonomous drones the size of bumble bees, then everything needs to be smaller – including the chips that control them. To solve this tiny problem, a team of engineers at MIT is working on a new computer chip that can be installed in a drone the size of a bottle cap, yet can capture images and navigate the device with a minimum of power.

Much of today's technological advances stem from what was once called the silicon chip revolution. The invention of the microchip that shrank in size and doubled in power about every 18 months made possible everything from the smartphone that fits in the palm of your hand yet outperforms supercomputers that once filled whole rooms, to complex robots with superhuman capabilities.

The irony is that in recent years drones have been shrinking at such a rate that a major limiting factor is the chip that serves as the drone's brains. True, they are much more powerful and versatile than ever before, but this power also means that the average chip is still relatively large (as in the size of thumbnail) and they consume up to 30 watts of electricity.

Led by Associate Professor of Aeronautics and Astronautics Sertac Karaman and associate professor in MIT's Department of Electrical Engineering and Computer Science Vivienne Sze, the MIT team is working on a more specialized chip that is much smaller, consumes less power, yet is up to the task of processing drone data.

Rather than the conventional approach of developing the chip first and the algorithms afterwards, the team has developed an approach they call "Navion," in which the algorithms for the chip and the chip itself are designed in tandem. One aspect of this approach involved paring down an existing algorithm responsible for determining the drone's awareness of its position in space, known as "ego-motion."

Different versions of this algorithm were installed in a very simple programmable chip called a field-programmable gate array (FPGA), which allowed the chip designers to produce the most efficient prototype that reduces the number of gates, and therefore the amount of power it requires, while maintaining an acceptable level of accuracy.

Using simulations from recorded drone data as well as motion-control flights, the team was able to produce a chip that conserves power by eliminating outside memory storage and processing data in such a way that it can erase any stored information the instant it's no longer needed. The result is a chip that could process 20 images a minute as it orients the drone in space, while consuming only 2 watts of power and two megabytes of memory.

The MIT team is currently working on improved algorithms and an application-specific integrated circuit that could one day get power consumption down to a matter of hundreds of milliwatts. Ultimately, the team hopes the technology will lead to Karaman says is "the smallest intelligent drone that can fly on its own." Such a chip could one day be used to make insect-sized drones for search-and-rescue missions as well as other applications.

"Imagine buying a bottle cap-sized drone that can integrate with your phone, and you can take it out and fit it in your palm," says Karman. "If you lift your hand up a little, it would sense that, and start to fly around and film you. Then you open your hand again and it would land on your palm, and you could upload that video to your phone and share it with others."

The research will be presented this week at the Robotics: Science and Systems conference at MIT.