Silicon Valley-based commercial drone company Kespry has demonstrated a prototype drone that utilizes an NVIDIA artificial intelligence technology to recognize objects and learn about its environment. The prototype, which is based on the Kesprey Drone System already being sold to the materials, mining, and construction industries, uses the NVIDIA Jetson TX1 module, which ups the device's intelligence by giving it the ability to run complex algorithms.
NVIDIA unveiled the credit card-sized Jetson TX1 module in San Francisco this week, saying that it would enable a new generation of smart, learning autonomous devices. The module boasts 1 terraflop of performance and is capable of handling machine learning, computer vision, GPU computing, and graphics generation on very low power usage.
NVIDIA claims the Jetson TX1 is the first embedded computer designed to process deep neural networks, which gives it the ability to not only recognize objects, but also interpret information. This machine learning is responsible for giving computers a host of new abilities, including interpreting conversational speech, navigating changing environments, and more.
While the module is set to find a variety of AI applications, the Kespry prototype demonstrates its potential for advancing the functionality of aerial and ground-based drones. The Kespry Drone System is currently used by businesses to capture aerial data and generate high-resolution maps and surveys of construction and mining sites, but Kespry CEO Paul Doersch says NVIDIA's technology will allow its drones to do much more.
"Today, Kespry customers already use aerial data gathered by our drones to calculate distances, sizes, and volumes," says Doersch. "With NVIDIA's new machine learning module, companies will be able to specifically identify construction vehicles, building materials and other structures, so they'll have even more relevant information to manage their job sites using commercial drones."
The technology also has obvious applications in the fields of robotics, and in search and rescue and security operations. Drones could autonomously perform search and rescue operations in remote locations, while drones or stationary camera systems could use facial recognition to identify subjects of interest in live feeds.
The Jetson TX1's specifications are:
- 1 teraflop GPU, 256-core NVIDIA Maxwell architecture-based
- 64-bit ARM A47 CPU
- 4K video encode/decode
- Support for 1400 megapixels/second camera
- 4GB LPDDR4 25.6 gigabit/second memory
- 16GB eMMC storage
- Wi-Fi and Bluetooth (802.11ac, 2x2)
- 1 GB Ethernet enabled
- Linux for Tegra OS
- Dimensions are 50 x 87 mm (1.9 x 3.4 in)
It is s available as both a module or a developer kit, with the latter available for pre-order now ahead of shipping in the next few weeks. The module will enter the market in early 2016 from distributors globally.
The prototype drone is demonstrated in the video below.
A useful additional feature would be the ability to know by wireless reporting the speeds of all traffic local to the parent vehicle's planned route and compare it with the average for the conditions and time of day etc. This would be an early indicator of potential problems and thus the need to slow down before being made to by meeting those problems themselves (Of course, a better solution is for the sat-nav to take them round those problems, if possible). It is going to take an extremely intelligent machine to cope with the all the demands of such a capability.
Also related to autonomous vehicles would be to have these devices 'looking' at the verges to each side and, if possible, assessing whether they can be safely driven on should the need arise - by meeting an extra-wide vehicle coming the other way on a narrow country lane, say.
What is necessary is for all autonomous vehicle design teams to at least arrive at a set of standards so that when they launch, they are able to provide the best performance possible and not be tied to some sales executive's notion of enforced product loyalty because their company's products have a unique system specification.