By Jason Wade | January 18, 2021

Bringing AI to Embedded Computing

Military embedded computing has always been about bringing intelligence to the edge. But the nature of that intelligence is about to change dramatically due to the rapid rise of embedded AI.

Embedded AI is an artificial intelligence capability that is deployed on a device or system such as a mission computer or similarly rugged computing solutions. If I had to choose just one application that will be dramatically improved and probably completely transformed by embedded AI it would be ISR.

Today, we are already collecting far more video data than can be fully exploited by human operators in real-time or even with today’s post-mission processing technology. AI is capable of handling very large volumes of data far more efficiently and accurately than humans or conventional processing systems.

AI technology takes hold

Embedded AI has become so pervasive in our daily lives that we regularly encounter it without giving it much thought. Take how a smartphone automatically unlocks when it recognizes its owner’s face. Or, how, if there is too much traffic on the way to work, we just ask Siri to open up Waze to find a better route. Similarly, advanced driver-safety features like Automatic Emergency Braking (AEB) and Lane Departure Warning kick in just when we need them to help avoid collisions. In each case, underlying AI technologies like voice recognition, facial and object detection, computer vision, turn-by-turn navigation, and autonomous driving have been around in various forms for decades.

We are now finally seeing a large variety of AI-based products in the marketplace due to the relatively recent availability of cost-effective AI-capable computer hardware. Certainly, the efforts of the automotive industry to deliver autonomous vehicles have accelerated the growth and commercial availability of embedded AI technologies. Today, there is a proliferation of new AI accelerator chips and a choice of open-source AI development platforms such as TensorFlow. Conditions are now perfect for an explosion of new AI-enabled capabilities in deployed military systems.

Mission-critical applications benefit from embedded AI

It’s almost like we’ve been on a binge stockpiling data, especially ISR video, in anticipation of the day we would eventually be able to fully exploit it. Essentially every mission collects ISR video. Cameras are deployed to capture as much imagery as possible, but the sheer volume of video can make it difficult to sift through to find meaningful new information.

In many cases, a mission requires real-time information and is dependent upon human operators to glean important details on-the-fly. However, it is well known that human operators quickly become fatigued and are notoriously poor at wading through video streams to find specific objects or infrequent events.

AI, on-the-other-hand, never tires and is capable of finding and tracking even difficult-to-spot or otherwise obscure objects, people, behaviors, or other changes in surveillance videos. Additionally, AI could be used to enable the camera only when specific elements or circumstances of interest occur so that only meaningful imagery is captured. This would significantly reduce storage requirements, thereby enabling much more efficient use of time, equipment, and space.

Equipping rugged computing solutions with AI

As a manufacturer of rugged computing solutions for military applications, our task is pretty clear: take the industry’s leading-edge processing capability and transform it to meet stringent military requirements. With the availability of COTS AI accelerators and open-source AI-development tools, suppliers like us can now offer military customers the opportunity to bring AI-capabilities to any mission.

I would venture to say that AI should be considered for every military ISR system. Customers have told us what they want most for their ISR applications is a system with a GPU in the smallest footprint possible. To that end, our team has focused on designing systems with a GPU-first perspective rather than a GPU-add-on approach. This has allowed us to provide enterprise-class processing performance in a system less than half the size and weight of a comparable rackmount server implementation.

To support AI-enabled ISR applications, we leverage advanced CUDA capabilities through NVIDIA® Quadro GV100 support in our compact ZM3 Computer. For customers looking for an embedded solution, we’ve developed our Insight Video Processing platform (releasing Spring 2021), a dedicated real-time video enhancement system that incorporates the NVIDIA Jetson TX2. The Jetson TX2 is built around an NVIDIA Pascal-family GPU and enables high performance AI at the edge.

If you would like to explore how AI can bring more value to your ISR applications, please contact us today.

Ready To Start Your Project?

Ready To Start Your Project?

Whether you are looking for an off-the-shelf product or a custom solution, we can help you meet your requirements. Contact us to speak with one of our solutions experts.

Let’s Talk