DJI on Monday unveiled a high-performance embedded computer running Ubuntu 14.04 for use in its Matrice 100 drone.
The Manifold supports CUDA, OpenCV and ROS, and it is compatible with third-party sensors. It lets developers connect a variety of devices — including infrared cameras, atmospheric research devices, and geographical surveying equipment — to the Matrice 100.
The Manifold collects and analyzes data live.
“Today, unless we’re talking Predator-class drones, sensor data has to be streamed to ground-based computers,” said Kevin Krewell, a principal analyst at Tirias Research.
“With the Manifold computer, some of that data can be processed on the drone, making it more autonomous,” he told LinuxInsider.
The Manifold can be ordered from DJI’s online store for US$500 now. Shipping will start Nov. 15.
The Matrice 100 drone costs about $3,300.
The Manifold’s Tech Specs
The Manifold measures 11 x 11 x 2.6 cm. and weighs less than 200 gm.
It can be mounted to the Matrice 100’s expansion bay.
The Manifold has an Nvidia Tegra K1 quad-core 4-Plus-1 ARM processor with 192 GPU cores. Maximum clock speed is 2.2 Ghz.
It also has a low-power Nvidia Kepler-based GeForce graphics processor, an image-signal processor and an ultra low-power audio processor.
Power consumption is about 15W. The Manifold has advanced power management, dynamic voltage and frequency scaling, and multiple clock and power domains.
The Manifold has 2 GB of DDR3L RAM and 16 GB of eMMC 4.51 storage, a microphone/headphone combined jack, two each of the USB Type A 3.0 and 2.0 ports, a micro-B USB connector that supports Force Recovery and Host modes, and a nonstandard USB 2.0 interface for drones.
Network connectivity is through 10/100/1000 Base-T Ethernet.
The Manifold has a half mini-PCIe expansion slot, a mini display HDMI connector, a UART connector, a micro SD card slot, and I/O expansion headers. It also has power, reset and recovery buttons. An external 14-26 V AC adapter is available as an option.
It runs the long-term support version of Ubuntu 14.04, which supports both multitouch devices and HiDPl screens.
Ubuntu LTS provides support and maintenance for five years from the release date.
The Manifold can be used in artificial intelligence applications such as deep learning and computer vision, DJI suggested.
“The Tegra K1 processor is designed to use its GPU for deep learning processing for applications such as automotive advanced driver assistance systems, and has significant capabilities to process data locally,” said Tirias’ Krewell.
That said, the Manifold’s physical and technical specs will limit the amount of AI processing it can perform.
“Autonomous cars can support much more powerful local recognition capabilities, but researchers are still using supercomputers for the heavy lifting of recognition training,” noted Paul Teich, a principal analyst at Tirias Research.
“Now cut the weight and power consumption by at least a factor of 10 each for drone-based automatic piloting,” Teich told LinuxInsider. “Drones will not be able to recognize the range of objects that a car can, and it will be much longer before they can do their own on-the-fly training, so to speak.”
The Manifold Is Not Alone
Nvidia already offers a development board similar to the Manifold in its Jetson TK1 development kit, Krewell pointed out.
The TK1 kit is a full-featured platform for Tegra K1 embedded applications.
It comes with a board support package and software stack that includes the CUDA Toolkit, OpenGL 4.4 drivers, and support for the OpenCV library for Tegra.
Devs can use the TK1 kit to create solutions for computer vision, as well as in the robotics, medicine, security and automotive fields.
“The technology [offered in the Manifold] is already available,” Krewell observed. “Manifold provides better integration for developers.”