The Cranky Admin
Peekaboo Drones and Edge Computing
The drones raise many concerns, including how to process data.
Thanks to computers, technology has a habit of moving from the lab to practical applications fairly quickly. While lab demos don't always make it -- there are no widespread commercial examples of Van Eck phreaking, for example -- all the tech we use in our datacenters today started as an idea in someone's lab. One such technology approaching commercialization is the ability to use radio signals to see through walls.
The original Wi-Vi MIT lab project was revealed in 2013, and used WiFi signals to see moving humans behind walls. It did this by tracking signal reflections; this is why people moving was a really big requirement.
Shortly thereafter, the same researchers were working on the next generation of this technology: WiTrack. WiTrack used a range of frequencies from 5.56Ghz to 7.25Ghz to basically create a radar system that could see through walls.
While the original researchers kept at it, the concept has received quite a bit of attention. Law enforcement agencies liked the idea. So do militaries. In fact, so many different groups are working on versions of this technology that it's safe to say that the Wi-Vi creators weren't the first to come up with the idea; they just got the most press.
Seeing through walls made the leap from lab to commercial product, and it's now even available in drone form.
I See You
While today's drones are an early iteration of the technology, there are a number of organizations working on fully-fledged peekaboo drones. The organizations pursuing peekaboo drones range from Silicon Valley tech startups to the military industrial complexes of multiple nations.
Peekaboo drones are quadcopters -- and other machines that can operate autonomously or semi-autonomously -- equipped with a variety of sensors. Specifically, peekaboo drones have sensors that allow them to "see" WiFi signals, mobile signals and so forth. Where vendors or operators are able to get commercial spectrum, Peekaboo drones are also likely to emit radio signals of their own, rather than rely on the crowded public spectrum.
Peekaboo drones are designed to work in swarms. The swarm's artificial intelligence (AI) isn't located on the peekaboo drone itself, but rather handled by a nearby server. Depending on the size of the swarm, the swarm controller can be portable enough to be carried by the swarm operator. Larger swarms, however, are predicted to have to rely on edge computing.
Peekaboo drones listen to the various frequencies they have sensors for and send that information back to the swarm controller. This makes radio spectrum management extremely important. All the prototype peekaboo drone swarms I've seen thus far use a form of mesh networking to keep the swarm working. The drones not only communicate with one another, but a few key drones -- sometimes called queens -- provide backhaul for the swarm to the drone controller.
All of this communication has to occur on frequencies other than the ones the swarm is using to observe the targets. The frequencies need to be clear enough that the drones can send all of their telemetry back without interference, and the drones need to handle all of this communication at distances that can sometimes be several kilometers from the swarm controller, without chewing into battery life.
The sensor data sent back by the drones is unfiltered. The drones don't have anywhere near enough compute power to process what they're collecting. The swarm controller can't process that data locally either; at today's compute power-to-size ratios, they're lucky if they can control the swarm with what they carry with them.
This means that any analysis of the actual sensor data from the swarm will need to be shipped off to a datacenter for processing. Several practical issues arise.
As one of the primary projected uses for peekaboo drones is to assist law enforcement with real-time situational awareness, latency is a serious consideration. While early solutions that give law enforcement a rough idea of how many people are where in a building will be helpful, ultimately the goal is to be able to perform detailed analysis of everyone in a building in real time.
To accomplish this, sophisticated Bulk Data Computational Analysis (BDCA) tools will be required. These tools will likely have to be located proximate to the swarm controller. Although this would seem like an ideal use case for edge computing, it may not always be the case.
While moving some of the swarm control logic off to the edge is unlikely to cause privacy concerns, it seems likely that there will be some pushback to the idea of law enforcement observing people with peekaboo drones and processing that data on servers likely owned by one of the major public cloud providers.
Additionally, peekaboo swarms may need to be deployed where no high-throughput, low-latency network connection exists. The best option may be to have a semi-truck full of servers and have the swarm controller parked in a van that can run a fiber optic connection to the datacenter-on-wheels designed to process the swarm's sensors.
Further complicating matters is that the most useful BDCA tools for performing analysis on peekaboo sensor data may not be something that is -- or even can be -- offered as an on-premises solution. The relevant BDCA tools offering enhanced machine vision, behavioral analysis and so forth may be proprietary offerings. Many of today's most powerful BDCA tools are.
Furthermore, it could be that those tools require specific custom hardware, such as custom ASICs, that only exist in the more central public clouds. This could put a damper on both the data privacy issues and the latency sensitivity.
It's highly unlikely that any individual peekaboo drone manufacturer can overcome all these issues on its own. Even the likes of Boeing would most likely have to partner with BDCA specialists like Google, Amazon or Microsoft to make a viable product that can meet all needs.
In the meantime, however, the development and commercialization of peekaboo drones promises to provide a much-needed, non-driverless-car use case for edge computing; one backed by nearly limitless funding, and that comes with a host of networking and spectrum problems for startups, regulators and network practitioners to solve.
Trevor Pott is a full-time nerd from Edmonton, Alberta, Canada. He splits his time between systems administration, technology writing, and consulting. As a consultant he helps Silicon Valley startups better understand systems administrators and how to sell to them.