New military automated airdrops will use image recognition instead of GPS | Hacker News Reader

New military automated airdrops will use image recognition instead of GPS

thestack.com

From the article: "The current implementation of JPADS relies on GPS tracking to guide unmanned drops to the correct coordinates, but any deviation from the drop target, which can be caused by wind or other environment factors, can result in the necessity for troops to venture into frontline territory to retrieve the delivery."

The image shows a parachute drop. I'm not sure how image recognition will help reduce the impact of wind or other environment factors. It seems like military GPS now in the sub-meter accuracy would be - by far - the best guidance system. Anyone have any insight as to how image recognition is better than GPS or even helps (although article implies replacement)?

It is harder to spoof image recognition guidance, but spoofing is not mentioned as an issue in the article.

Ah.. I think you might be onto something. An image recognition approach probably does mitigate gps spoofing and jamming attacks. And perhaps they are thinking about china's anti satellite missiles that they tested a few years back.

Is it really? The military has been using camo to spoof image recognition since forever.

But I guess if this is for drops then camo wouldn't apply?

> Is it really? The military has been using camo to spoof image recognition since forever.

I don't think so. Camouflage has been used to hide things in the terrain for a long time, but I don't think it's ever been used impersonate a different area's terrain (which I think you'd have to do to "spoof" in this context).

Edit: Actually, now that I think about it, I might be wrong. Wartime blackouts could probably be thought of a kind of spoofing camouflage, especially if they put decoy lights in an area away from the city or left lights on in less-critical areas.

https://en.wikipedia.org/wiki/Blackout_%28wartime%29

Was used in Area 51: http://news.nationalgeographic.com/news/2011/05/110520-area-...

And while not images per se (though they did use blow up equipment), there's the Ghost Army:

https://en.wikipedia.org/wiki/Ghost_Army

Edit: To clarify, the Ghost Army used inflatables not to hide, but to mimick and deceive, but the concept is similar.

There is the unspoken assumption here that several methods are used to guide airdrops to the proper location. All of them coupled together are strong against any attack which targets one. Something else is that specifically there would most likely be markers (infrared fireflies, colored flares, smoke, etc) that would be used to signal the drone in the first place. The infrared spectrum is still phenomenally useful because of the barrier to entry the technology has for utilizing it. If you see a flashing infrared signal, it is almost a certainty it's US or "allied" forces. Kinda tangential but interesting.

I agree. I can't think of a better or more reliable way to place one object at/near another. If they both have GPS units, the hard part should be steering the parachuting drop not knowing where to steer to, right?

1) GPS is a weak signal and can be jammed or spoofed.

2) Your satellite imagery might not be accurately aligned to the coordinated system. If you want to land on a little flat spot, then doing it visually will work even with misaligned imagery.

I'm sure the NRO can source some really accurate satellite and drone based terrain imagery if needed.

Maybe, but the cost in that infrastructure is the bureaucracy of getting that data to the machine doing the dropping. There is a huge walled fortress surrounding imagery acquisition and analysis that would make tunneling that information to the airplane/drone in a timely manner very hard in its current form.

You don't need to use a coordinate system at all with scene match guidance.

You can use what ever imaging you have since the guidance system calculates the position based on the position of known points in the scene.

If you want to drop something you just pick the spot visually on the map and it will work better than gps.

Visual guidance and INS are considerably more accurate than GPS when it comes to terminal guidance.

My guess is the military is using image recognition in ADDITION to gps like you suggest. But I too don't see how that is going to help with the wind. Also I would assume you would drop the package several miles behind the front lines to avoid that sort of problem (Assuming the allied forces are not surrounded). Maybe they are referring to placing the camera's directly on the payload but I still don't see why that would be better then gps.

Thinking about it, i wonder if it is about "intelligently" locating an alternate place to touch down when wind etc makes it impossible to reach the GPS location.

Could be that right now it dumbly steers towards the location no matter if that brings it smack into a hillside or similar, thus ending up in unpredictable and possibly hostile locations.

I find it hard to believe they can pull this off, but it is theoretically possible that they could visually look at clouds vs trees and trash blowing about to correct for the difference between surface winds and winds aloft, which sometimes can be quite a difference.

If you somehow magically knew the wind vector at various heights (visually, I guess, although a mesh network of cooperating air drop parachutes would work), even relatively slow actuators could very accurately pinpoint land.

Another "I can't believe it, but maybe..." is GPS doesn't have a plot of every tree or vehicle and visual could be programmed, in theory, to do something ridiculous like thread the needle of individual trees and bushes to land next to visually identified military vehicles, where ever they may be parked, or even crazier, supply on the move vehicles, so if we're hitting the ground in 30 seconds and that APC is moving at 25 MPH to the east then I want to land 20 feet away from where it'll be in 30 seconds while missing the nearby trees...

Another strange visual landing idea... regardless of GPS jamming, using narrow beam optical signalling make a mesh network of dropping pallets and visually coordinate landing such that you avoid all trees vehicles and people while landing in a super convenient straight line exactly 50 feet apart regardless of how/when the pallets were kicked off the plane.

>... it is theoretically possible that they could visually look at clouds vs trees and trash blowing about to correct for the difference between surface winds and winds aloft

They[0] achieved remarkable drop retrieval rates in Vietnam, often over 95% and sometimes 100%, by doing just that; having an observer on the ground report the surface conditions. The navigator had a very stressful few minutes on the run-in adjusting his previously-calculated ballistics.

A easier and superior technique was high-speed dropping using reefed parachutes. The payload would descend at 100 metres / second, basically immune to crosswinds. This wasn't suitable for medicines, fuses or other fragile payloads but was fine for munitions, rations, rice and tins of peaches ( a tonne of which once squashed a jeep ).

HALO dropping was also used but whilst being gentler on the payload it was less accurate and more susceptible to rigging failures.

The Caribou squadrons adopted a technique later used by Blackwater in Afghanistan; fly directly over the drop-zone at 50ft, pitch-up and kick the pallets right out the back.

[0] 374th Air Wing and later the Vietnamese airlift squadrons who were trained in the same techniques

Imagine semi-AI parachutes and pallets that measured their trajectory as they landed and fed that data back to the parachutes a few seconds behind them such that the landing precision increased automagically when you drop multiple pallets. Even better prioritize by contents so stuff thats tough and easy to recover by hand lands first as the guinea pig and signals back to all the other soon to arrive pallets whereas you can nearly guarantee the mortar ammo lands within 10 feet of the mortar weapon. Why you could even kick the mortar and mortar rounds off different aircraft and theoretically they'd land right next to each other in formation, whereever they happened to land.

Speaking of guys on the ground using some of those "invisible" IR chemlights and some smoke grenades the soldiers on the ground could visually signal conditions back to the parachutes while the parachutes are falling... This formation of flares on the ground or whatever means hot LZ so use dense hot LZ pattern, this formation means crazy civilians are rushing the field about to get squashed so disperse, this pattern means the LZ has just been overrun so fire the thermite grenades while you're still in the air...

Another strange idea, do the helo LZ thing where you try to secure three LZs to throw the enemy off, and only after the parachutes are already falling, throw smoke and the parachutes all formation fly to the pre-determined LZ. So instead of the enemy having an hour to react or tens of minutes, the troops will be resupplied a couple minutes at most after its obvious they're getting supplies.

GPS in wartime is a problem as it can be spoofed or jammed. This is how some think Iran managed to capture an RQ-170 drone.

The military is very motivated to find ways to be less reliant on GPS and other easy to spoof technologies.

There's a better article here:

http://www.defenseone.com/technology/2016/01/army-testing-ro...

>GPS “makes us vulnerable to attacks, it is impossible to use in the valleys of Afghanistan or in a big city [where signals are blocked], or in places where the signal is poor.” “At DoD, we worry about enemies jamming GPS signals.”

>“Threats to military GPS have evolved and improved at a rapid pace — from a proliferation of small-scale commercial jamming devices that can readily be purchased on eBay to large-scale military anti-access/area-denial (A2/AD) capabilities.”

"which is aimed at reducing the 3000 casualties which in 2007 resulted from inexact air-drops in militarily sensitive environments such as Afghanistan."

Is that saying 3000 casualties just in 2007? That number is a lot higher than I would have expected it to be. I didn't realize this was such a problem.

I believe that number includes casualties caused from retrieving the package, but it still seems high to me too

There were 7,700 people killed in the war in Afghanistan in 2007. Forty percent were due to inexact airdrops?

https://en.wikipedia.org/wiki/Civilian_casualties_in_the_war...

Casualty is a military word that includes the sick and the wounded. Doesn't mean they died.

Another part of the problem was using the same colour packaging for food-aid air drops and cluster munitions. About 10% of cluster bomblets don't explode.

https://www.hrw.org/legacy/backgrounder/arms/cluster-bck1031...

http://www.landmineclearance.org/page2.html

http://www.rawa.org/cluster2.htm

Yeah, not only that, but they look like big soda cans. If there's one thing little kids have eyes for, there it is.

    But most telling is the political and military calculation that 
    underlies the Afghan civilian bloodletting. 
    "Close air support" bomb attacks called in by ground forces - 
    which rose from 176 in 2005 to 2,926 in 2007 and are now the 
    US tactic of choice - are between four and 10 times as deadly 
    for Afghan civilians as ground attacks, the figures show, and 
    air strikes now account for 80% of those killed by the 
    occupation forces.

From http://www.theguardian.com/commentisfree/2008/oct/16/afghani...

Reminds me of TERCOM's predecessor, DSMAC - https://en.wikipedia.org/wiki/TERCOM#DSMAC

Yeah, totally thinking of missile terminal guidance.

I guess fear of jamming, capture, and the need to keep system costs low is why they're not going for a mapping radar approach instead as that would get them all-weather capabilities.

In addition, a passive system is useful for covert as well as overt drops. An active system isn't.

Better link that describes the other concerns about GPS that were brought up and describes in more detail how the system works: http://www.defenseone.com/technology/2016/01/army-testing-ro...

> The testing of a GPS-free version of JPADS is yet another example of efforts to develop alternatives to satellite guidance. “GPS is very expensive to launch and operate,” Defense Secretary Ash Carter said during an April 24 podcast from Silicon Valley, according to National Defense magazine. The global positioning system, he said, “makes us vulnerable to attacks, it is impossible to use in the valleys of Afghanistan or in a big city [where signals are blocked], or in places where the signal is poor.” “At DoD, we worry about enemies jamming GPS signals.” Army Maj. Christopher Brown, who helps run the service’s efforts in position, navigation and timing, or PNT, told C4ISR & Networks last year. “Threats to military GPS have evolved and improved at a rapid pace — from a proliferation of small-scale commercial jamming devices that can readily be purchased on eBay to large-scale military anti-access/area-denial (A2/AD) capabilities.”

So the real benefits here are: works in mountains, and future-proofs against better GPS jamming.

Iran managed to "hack" GPS to steer a state of the art American drone to land safely at a location of their choice over four years ago.

https://en.wikipedia.org/wiki/Iran%E2%80%93U.S._RQ-170_incid...

I would say that GPS's vulnerability is a current problem rather than a future one.

Time to bring back camouflage cities[1] and ghost armies[2] then. [1] http://www.lockheedmartin.com/us/100years/stories/camouflage... [2] https://en.wikipedia.org/wiki/Ghost_Army

My bad; now I know what an air drop is, and it doesn't involve blowing anything up. Still, the above links are fascinating.

But Ghost, uh, Terrain can still make it possible to steal airdrop supplies by confusing the camera en route to its destination.

I've some of the self-driving systems are relying on remembered images of highways as much as lidar or radar. A memory of roads near your residence and workplace is useful.

This might save lives in Modern Warfare's Drop Zone mode.

Reply on news.ycombinator.com