MIT researchers designed a microchip last year designed to help micro-drone navigation —— today, their chips have undergone an upgrade.Not only is it smaller, it also consumes less energy.
The team was jointly led by Vivienne Sze, Assistant Professor of the EECS Department of Electrical Engineering and Computer Science, and Sertac Karaman, Assistant Professor of Aerospace and Aerospace. From scratch, they customized this chip with low power consumption, small size, and increased processing speed.
This new chip is named "Navion" and is only 20 square millimeters in size, which is about the size of the Lego doll's footprint. It consumes only 24 milliwatts and is almost one-thousandth of the energy consumed by an ordinary light bulb.
Although Navion consumes minimal energy, it can process real-time images of 171 frames per second and perform inertial measurements, both of which are used to determine its own position in the air. Researchers say the chip can be integrated into finger-sized "micro-drones" to help drones navigate, especially in remote, unmanned areas where global positioning satellite data is not available.
This chip can also be used for small robots or other mechanical devices that require long time navigation but limited energy storage.
“I can imagine using this kind of chip for low-energy robots that run on a battery for months, such as nail-sized flapping aircraft, or devices that are lighter than air like a weather balloon."Mr. Karaman, a member of the MIT Information and Decision Systems Laboratory and the Institute for Data, Systems and Social Research, said, "Imagine its use in medical devices. For example, if you swallow a small pill, it can Rely on a very small battery for smart navigation so it won't overheat in your body. The chips we are developing will play a big role in these areas. ”
In the past few years, many research teams have designed and manufactured miniature hand-sized drones. The scientists envision that such small drones can fly around and take photographs of the surrounding environment. After completing the mission, the drone can be returned to the user's palm and easily stored.
However, the batteries that a palm-sized drone can carry are very limited, and most of the power has to be used for motor work. In addition, only very little power is used for other basic operations, such as navigation or state estimation, that is, to determine its own position in the air.
“In traditional robotics, we use an off-the-shelf computer and implement a (state estimation) algorithm on it, because we don't usually have to worry about energy limits."Karaman explained," but in every project that needs to miniaturize low-power applications, we must think about the challenges faced in programming design in a completely different way. ”
In previous studies by Sze and Karaman, they have tried to integrate algorithms and hardware into the chips they designed to try to solve this problem. Their initial design utilizes an FPGA, which is a commercial hardware platform that can be configured according to the application. In this case, the chip can perform its own state estimation at 2 watts, which is much lower than the 10-30 watt required for large drones. Despite this, the power consumption of 2 watts is still greater than the battery power that can be carried by ordinary micro-unmanned aerial vehicles. The average micro-drone battery power is only about 100 milliwatts.
To make the chip smaller and use less energy, MIT researchers decided to design a new chip from scratch instead of improving existing designs. Sze commented: "In this way, we have greater flexibility in the design of the chip." ”
Exert yourself in the world
In order to reduce the energy consumption of the chip, the research team's new design minimizes the amount of data stored in the chip at any given time, including stored images and inertial measurement data. The new design also optimizes the data path in the chip.
“We compress any image that needs to be stored on the chip temporarily, so it can take up less space.” Sze said. The team of researchers at the ’s Research Laboratory of Electronics, led by Sze, also reduced unnecessary operations on the chip, such as some zero correlation and zero-computing operations. The researchers found a This method can skip any calculation that involves zero. Sze commented: "This allows us to avoid processing and storing these zeros, so we can reduce a lot of unnecessary storage and calculation cycles, thereby reducing chip size and power consumption, and improve the chip processing speed. ”
With this latest design, researchers reduced the chip's storage space from 2MB to about 0.8MB. The researchers tested the new chip using a drone flight database previously collected in different environments (such as workplaces, warehouses, etc.).
"We not only allow the chip to use less energy and operate faster, but also allow it to flexibly adapt to different environments to further save energy." Sze said, "The key lies in finding a balance between flexible & rsquo; and ‘high efficiency ’. "The chip can also be reconfigured to support different camera and inertial measurement unit (IMU) sensors.
Through testing, the researchers found that they can reduce the energy consumption of the chip from the previous 2 watts to 24 milliwatts. 24 milliwatts can support the chip to process 171 frames per second —& mdash; this efficiency is higher than the efficiency of the drone when collecting the database.
Researchers plan to have a miniature car equipped with chips to display this new design. They want to display real-time video of a car camera from a single screen, showing the chip's calculated airborne position, and the power consumed by this task.Since then, the team plans to test the chip on a real drone and eventually test it on a micro drone.