From pizzas being flown in to pharmacies that deliver medicine, the idea of connecting the warehouse to the end customer using drone delivery is the one hype from online retailers that gets the most attention.

There is even talk of automating delivery, which makes a lot of sense, since the weakest link is the drone pilot today. Whatever the economic reality, and I’m sure that there is a case for drones, we need to look at the implications of gear failing before it’s too late to stop my worst-case scenario—that of furniture store delivering a sofa by drone that loses the sofa over a populated area.

Wait, you say, who’s talking sofa delivery? Well, what happens if a car piston is dropped on your head from 60 metres or a big screen TV hits your windscreen? That is just as bad as the sofa and quite realistic as drone payloads go. But is it going to happen? Sadly, it is a certainty. In fact, I’m surprised it hasn’t already happened.

It’s a matter of simple statistics. Mean-time-between-failure applies to drones as much as any other electronic device. Statistically, a portion of any population of drones is going to die today. Let’s take a stab at how many that will be.

The typical drone has four motors. Each of these is good for maybe 10,000 hours of use, on average, so that’s 400 failures in one million hours of use. Statistically, that means a motor will die on average every 2,500 hours, or roughly 100 days of continuous use. Now, let’s add in the control electronics, which is about a cellphone’s worth of smarts. That portion of the drone has an MTBF of roughly 50,000 hours, or 20 failures per million hours, the camera adds roughly 20 more failures and the batteries add another 20 failures.

Add these up and we get 460 failures per million hours which translates into one every 2,000 hours. Let’s deploy our drone fleet of 20,000 drones (it will take that and some for Amazon). That average failure rate of once every 2,000 hours, becomes 10 failures per hour.

That’s right, 10 drones an hour will crash and burn (statistically). The good news is that much of the damage will be to the drones and their cargo and not much else, but there’s no way to ignore the fact that drones will fly over people for at least part of their trip.

Wait, though, because the news actually is worse. I looked at failures in a carefully handled environment. Anyone who has been near a warehouse knows the loaders have contests for how far a box marked ‘Fragile” can be tossed. Moreover, drones will see all kinds of weather. All of this makes the above failure rate optimistic.

So, drones won’t be reliable enough for the job. Compared with even the worst delivery service, the lost deliveries will kill their reputation, while the insurance companies will avoid the risk. I have not seen anywhere near the level of discussion that the reliability topic deserves. It’s all lost in the gee-whiz of drones displacing pizza deliverers and UPS drivers.

Can anything be done to save the mess? Well, the largest contributor to failure is the motor, so let’s add some redundancy and use six motors per drone, so that a single motor failure still allows the drone to fly.

That’s an easy fix. The rest of the failure modes need a technique called built-in redundancy to resolve any risk. Instead of a single controller, there needs to be a way for a second, back-up unit to kick in and control the drone, irrespective of whether the pilot is human or an AWS virtual instance. Redundancy is more than just two sets of electronics. There has to be hardware and software enough to detect the failure and switchover automatically.

There are other subtleties with drone reliability. Even with redundancy, failures will happen. It really is important that we be protected from this, so a fail-safe such as a parachute or helium balloons to soften the crash may be needed and reporting a crash should be built into the drone electronics.

With automated flight control, where to land becomes an issue. Dropping a parcel in with four pit bulls may not be the best idea, for instance. In fact, without a designated landing area, the risk of the parcel and/or the drone being damaged is too high. So, “X” marks the spot. That might be great for Target stores.

To sum this up, we are still starry-eyed about drones and need to address the practicalities of flight safety real soon. Unless we figure out redundancy and stability, delivery by drone will just be a pipedream, with a history of failed projects and, sadly, spectacular drone crashes. We CAN do better. If we do this right, drones could solve the problem of home delivery in an online retailing world.

First published by EBN.