>Flying blind to a preprogrammed GPS location is simply impossible in the real world.

I didn't intend to rule out the need for clever self-flying technology on the part of Amazon.

But comparing the challenges that face an airplane that carries people at very high speeds and altitudes to a low altitude slow-moving drone is pretty silly.

Google's self driving car does this at road speed limits. There's no reason a drone doing 20 MPH with an extra degree of freedom couldn't do it as well.

>Also, remember that GPS isn't quite enough to drop a package right on your doorstep. You need much more than that.

>> at which point control is handed to a pilot for the landing & package drop

>> at which point control is handed to a pilot for the landing & package drop

Good, how?

Do you think 4G can allow you to do that? With enough latency?

Flying military drones is a very complicated matter.

Today civilian drones are flown with visual contact and at close range.

So, no.

>Good, how?

Well, if we assume that the drone is pilotless up to the point when the drone needs to land:

* the drone flies to the GPS point in question and hovers pending instruction

* the drone takes a picture of the ground area and sends it via 4G to a controller

* the controller identifies a landing point and transmits it back to the drone

* the drone uses its obstacle avoidance systems to land, drop the package, and take back off for a return to Amazon's center.

This is just one idea. Perhaps Amazon provides you with a beacon for the drone, or you get a big QR code for the drone to target your house & they don't need pilots at all. Who knows!

Either way, it seems to me that the most interesting thing would be the collision avoidance systems for automated flight. But considering Google can do this with a car on highways, I'd imagine that a slow moving drone that can also operate along the Z axis would be easier.

> Flying military drones is a very complicated matter.

I'm sure it is - but military drones have an incredibly different mission; comparing them to a slow package delivery drone isn't really worthwhile. Hell, according to Wiki the Predator drone's stall speed is 60 mph - Amazon drones sound like they will be much slower, at least at first.

Do I misunderstand this technology? I think the fly to GPS waypoints thing is done by these guys: http://copter.ardupilot.com/

See this: http://copter.ardupilot.com/wiki/planning-an-apmcopter-missi...

The flight is manually pre-planned. You also need to consider obstacles (and other aircraft).

High latency's hardly going to be the end of the world for a slow-moving, low-mass n-copter that can just sit where it is during a lag spike.

I think the point the parent was making was lack of real-time feedback for the pilot.

Yes, my point was that such a device is going to be relatively easy to fly remotely even over a balky, high-latency connection. (Except perhaps in high winds, but high winds are likely to ground these things anyway.)

Airplanes are almost automated and pilots seem to be just there as a safety backup: http://uk.reuters.com/article/2013/10/08/oukoe-uk-britain-pi...

No, air traffic is not 'almost automated', and pilots are not 'just there as a safety backup'. The only part of flying that has been automated is avionics, ie: taking off, flying the aircraft along a pre-set flight path, and landing it. While this may seem this is 'all you need to automate away pilots', it completely ignores these nasty few percent of tail risk where things turn out not to follow the exact preconditions at the time of the flight plan. Like weather, technical problems (in the plane or on the ground), delays, airspace congestion, diversions, whatever. In each and every one of such cases it's still the pilot that has to decide how to handle the situation and either fly the plane itself, or reprogram the autopilot.

People always seem to think planes fly 'autonomously', but they don't. They can fly 'automatic' maybe 90% of the time, but they don't fly themselves.

>In each and every one of such cases it's still the pilot that has to decide how to handle the situation and either fly the plane itself, or reprogram the autopilot.

So far I've heard these drones won't work because existing aircraft can't do this, or because military drones can't do that.

But no one seems to consider that these drones don't fulfill the same role as a Predator drone or a jumbo jet, so many of these objections are incredibly odd.

I remember seeing a link on HN several weeks ago about a flying robot that was designed to bump into walls and reorient itself. What if we had an army of package delivering robots that flew at 5 MPH, did best-case collision avoidance, and if they bumped into the side of a brick wall it's not a huge deal?

If an Amazon drone runs into bad weather it could be programmed to land on the nearest flat roof. If unexpected airspace congestion is detected it could just turn back to home. Or hover in place until things resolve. Or(...)

We aren't trying to safely land an airplane full of humans at 100 knots, or deliver a Hellfire missile on a moving target. We are flying a lightweight, slow moving drone with a tiny package. The considerations are completely different.

>> So far I've heard these drones won't work because existing aircraft can't do this, or because military drones can't do that.

That's not the reason why I think this Amazon drone thing is just a way to generate some media hype, I can think of many other reasons why the idea is silly and won't work. You can find most if not all of them in other peoples posts in the various Amazon drone articles, I don't have to list them here.

My main point was not that delivery drones are impossible because planes or military drones still need pilots, I just wanted to point out that 'self-flying planes' are so often used as proof for other types of autonomous vehicles, which is based on the false premise that planes are autonomous.

90% is probably good enough for Amazon package delivery. The other 10% they send out another one. The cost of this is included in the fast-delivery surcharge. Unique and living products are not eligible for air delivery.

Devices that do this already exist and have done since 1983: Tomahawk cruise missiles. Huge areas of military technology have been developed for precisely delivering explosives; now they are being repurposed for non-harmful use.

GPS is certainly enough to fly it to the visual vicinity of your house. I can imagine needing to supply a landing marker (large QR code?) to tell it exactly where to land.

A QR code is a nice idea. Another solution could be to take a picture or a short video of the place you want the drone to leave the package. Machine vision technology is easily sophisticated enough to translate this into an accurate 3D model which could be correlated with known GPS information.

This kind of process will probably happening anyway for hazard avoidance.

It's pretty easy for a drone to fly higher than almost all the obstacles you just listed. Skyscrapers and smokestacks are going to be a concern, but those don't move and it's easy to do offline route planning to avoid those. Construction cranes and other temporary obstacles are a problem, but some rudimentary sensing and fallback to human control should solve that.

As someone who makes delivery robots for hospitals, let me say that I envy navigation problems of flying drones. I'd seriously worry about power and range and weather and how you recover a drone that has a problem and crashes, but not navigation in general.

Amazon doesn't need this to use drones for 100% of deliveries. But if they pick enough constraints, could it work?

Assume it will never work in NYC, but could it work in Albuquerque, New Mexico? Assume it won't fly if the winds are over mph. Assume that only human operated landings are performed the first time to scope the landing zone, with future approach and landings to be automated. Assume that anything unexpected causes the delivery to get canceled.

In the 60 Minutes interview, Bezos was optimistic but more cautious about when it would be ready -- more like 2018+. Also, the earlier part of the segment was about how it took Amazon Fresh 5 years to expand grocery delivery to its second city. Bezos, of all tech companies, knows the value of staying with a problem for the long term. If I had to bet if Amazon could figure this out over the next 50 years, I would say yes. Sooner is a bonus.

>In the real world, you can't "fly to this GPS point". If this were possible, we would have automated airplanes and pilots would be out of their jobs.

These things don't follow. We actually do have automated airplanes and pilots jobs are really just there for passenger comfort, but that is because Autoland and autopilot utilize ILS' and other sensor data besides GPS to make it precise enough.

So while your point about GPS alone being insufficient for precision delivery is correct, the overall thought that we can't do precision delivery is misguided.

While it is hard, impossible is a stretch. You can buy drones with autopilot that will land at a point as long as you have 10 ft of clear space. A friend yesterday was using one he'd received for $3000 to film some cliff tops while we were walking the edge. After shutting off the remote it can fly back to where it was turned on so you don't have to carry it. Bear in mind this is with tech that's a few years old.

Yes, as long as there isn't a crane, net, tilted lamppost, kite, human, dog, or any other unplanned obstacle.

Come on, people. Being able to draw a trajectory isn't the same as moving a physical object through space. And moving a drone to a predetermined GPS location in ideal conditions while you are watching it and when you know there is nothing around has nothing to do with day-to-day deliveries using drones.

So you believe sensor based obstacle avoidance is an impossible problem in robotics? Second, unrelated question, do you think drones need to complete 100% of their scheduled tasks or do you think it's acceptable to fail a small fraction of the time?

I know, but with the Raspberry Pi's and similar tech we can't be that far away from something that's technically feasible. We also don't know what Amazon has been working on behind closed doors. They are known for vast technical prowess after all.

I would disagree with it being impossible, especially in the drone delivery business. Definitely hard but not impossible. The drone leave a known location to go to another location using a direct path. It is a point A to point B algorithm. Amazon could build or use a database of information regarding obstacles and traffic, to program and test well ahead of time destination between A (the warehouse) to B (i.e. every doorstep in a 30 minutes radius) and back to A. GPS is often accurate within 1 meter which should be enough for door delivery (http://www.gps.gov/systems/gps/performance/accuracy/histogra...). I agree that humans will be involved in monitoring the drones 24/7.

Did you watch this?

http://www.youtube.com/watch?v=geqip_0Vjec

Do you see all those cameras surrounding that drone? You don't have those in the real world.