Decisions on the fly

– The challenge with AI in small, has always been how to  get engineer/maker to get started. Many Learning is out of reach  without to robots and drones. In many cases mobile devices such as drones, the level of computing needed to do advanced AI techniques like Deep Learning into a small size and power envelope. To address this, NVIDIA released the Jetson TK1 Developer Kit in 2014. With support for computer vision and Deep Learning acceleration developers were able to get started with some of these techniques. Several great products were launched on this architecture, including Percepto and the DJI Manifold. Now we have the Jetson TX1, which offers tremendous acceleration for AI in a 10W package. It’s a great product for tasks like automated aerial inspection, which is what our partners Aerialtronics and Neurala handle, and with warehouse inventory management from Intelligent Flying Machines and Squadrone Systems, says Clayton.

– In addition to performance improvements, NVIDIA has also tried to make Deep Learning more accessible. For instance, Make magazine recently gave step-by- step instructions on how to quickly build a deep learning demo on a Jetson TX1 Developer Kit, which highlights how easy it is for a cleveR companies may feel that Deeprealizing that they already have the resources to get started  in-house, Clayton explains.

If looking back, when and why did NVIDIA see the potential of your technology in the UAV marked?

• – UAVs are unique because they can go where people and other devices cannot.

– UAVs are unique because they can go where people and other devices cannot. While they offer substantial promise to solve a variety of problems, they are limited by their reliance on GPS-based navigation  and the availability of qualified  pilots. Given NVIDIA’s history in self-driving cars it was only natural for us to explore other areas where this technology could be applied, says Clayton.

Ground-robots and UAVs were two obvious applications.

– One of the benefits of NVIDIA’s technology is that the same core architecture is used in all of its high performance computing products, from the Titan at Oak Ridge National Labs, all the way down to Jetson for robots and UAVs. When someone develops on NVIDIA for a workstation, car, or other form factor, it’s easy to transfer that work it’s only a recompile, says Clayton.

Looking forward, where are we going and what can we see on the  UAV marked in 2017? And if looking further ahead of us, which trends do you see ahead us?

– Now’s the time for companies to take a serious look at their AI strategy not just for UAVs, but across all industries. Those that don’t may be at risk of being left behind.

– 2017 is set to be a very important year in the UAV space. Some of NVIDIA’s customers are already deploying advanced AI to the field for tasks such as industrial inspection and warehouse management. We also see many additional opportunities. In agriculture, package delivery, emergency response, and security. Now’s the time for companies to take a serious look at their AI strategy not just for UAVs, but across all industries. Those that don’t may be at risk of being left behind. The good news is that techniques like Deep Learning are very accessible and most companies already have the resources in house to get  started, says Clayton.

NVIDIA’s Deep Learning page is a great first step: https://developer. nvidia.com/deep-learning

If you create a vision of what a drone can you with the DL / AI in  a few years – what would it look like if your expectations were right?

– Wouldn’t it be better if the pilot could just bring the drone out to the site, press a button, and have the drone handle the entire task? It’s faster, more efficient, and safer. Companies like Aerialtronics are working on this today and my expectation is that we’ll see massive deployments of fully automated systems down  the road.

– NVIDIA sees tremendous opportunity to improve productivity, reduce task times, and take people out of harm’s way using AI technology on UAVs. One example is in industrial inspection. Today, if an operator wants to inspect a cell tower or a wind turbine, a pilot with a drone needs to be sent out to the site to fly around and capture footage that’s then brought back to the lab. If anything’s missed, the pilot must be sent back out to complete the job. Then an expert reviews the entire footage looking for problem areas and schedules the maintenance to be performed. It’s crazy! Not only is it inefficient, it can also be dangerous since the pilot can potentially fly into the tower or another obstacle. Wouldn’t it be better if the pilot could just bring the drone out to the site, press a button, and have the drone handle the entire task? It’s faster, more efficient, and safer. Companies like Aerialtronics are working on this today and my expectation is that we’ll see massive deployments of fully automated systems down  the road.

Other areas where AI can provide significant value include agriculture, package delivery, emergency response, and security. Now with Jetson all of this is within reach.

What are the limitations for the technology today?

• It’s no secret that AI is a computation-heavy activity. Jetson TX1 offers an incredible amount of performance in a small form factor. While it’s really pushing the boundaries of computation, it cannot solve all problems. NVIDIA, its partners, and customers need to continue to innovate and test the limits of what’s possible to make a real impact.
• The good news is that those things are already happening with NVIDIA pushing forward on new architectures and capabilities. In addition, the NVIDIA and Jetson ecosystem are making it easier than ever to take advantage of AI for UAVs and other robots. It’s a very exciting space to be in right now.
• Now’s the time for companies to take a serious look at their AI strategy – not just for UAVs, but across all industries. Those that don’t may be at risk of being left behind.
• Want to know more about DL / AI. Check out these companies, they have created drones or develop software found in drones using Jetson TZX1 or TK1 Developer kit.

https://developer.nvidia.com/deep-learning https://developer.nvidia.com/embedded/learn/success-stories,

Aerialtronics: Altura Zenith is one of the first commercial drones to use AI technology to visually inspect buildings, cell towers, wind turbines and more. All real-time processing is done onboard the drone.

Birds.ai: Computer vision software for precision agriculture and inspec- tion, enabling drones to count cattle, crops and more. Customers can get a bird’s eye view of all their assets through aerial imagery.

DJI Manifold: Developer platform that can turn drones into intelligent flying robots that can perform complex computing tasks and advanced image processing.

Enroute: A scalable platform for autonomous drones and unmanned ground vehicles (UGVs) that uses deep learning for search and rescue, infrastructure maintenance and land surveys.

Hexo+: An intelligent drone that can capture footage autonomously, making aerial filming possible in the most remote locations. This self-flying camera follows and films people from takeoff until landing.

• IFM: An autonomous drone that can find its way around indoor spacesand geared towards warehouses. It’s smart, light and agile enough to zip through tight spaces and smart enough to accurately track inventory.
• Neurala: Deep learning software, Brains for Bots, runs in real time on a drone. With Neurala software, a drone can learn objects, scenes, people or obstacles.

• Parrot S.L.A.M. Dunk: An open development kit for the design of advanced applications for autonomous navigation, obstacle avoidance, indoor navigation and 3D mapping for drones and other robotic platforms in environments with multiple barriers and where GPS signals are not available.
• Percepto: Advanced vision capabilities for autonomous drones. The platform was recently updated to enable commercial drones to connect to a 4G/LTE network to handle long distance security, safety and inspection tasks autonomously.
• Squadrone Systems: Real-time data collection and data analytics for logistics, site exploration and surveillance. These autonomous flying drones can scan items in bins and recognize misplaced items.
• StereoLabs: The world’s first high-definition 3D camera for depth-sens- ing that will create a detailed three-dimensional map in real-time.
• Teal: The world’s fastest commercial drone with an onboard flight control Follow-Me feature, which allows the drone to autonomously follow a person.