Hacker News Reader

Comment by runjake | Hacker News Reader

Just keep in mind, this is a science project. It's going to be a long, long time (or never) before this is safe from enemy exploitation. I can think of all kinds of wacky ways to crash one of these things.

Autonomy is fine for high-flying, loitering aircraft, not so fine for low-altitude flight relying on easily-exploitable terrain avoidance & positioning technologies.

View on HN

Already fielded: http://en.wikipedia.org/wiki/Kaman_K-MAX#Unmanned_remote_con... I presume the difference is that the K-Max is not doing terrain avoidance using laser ranging (undoubtedly it has a preprogrammed 3D flight profile).

"In December 2010 NAVAIR awarded a $46 million contract to Kaman for two aircraft, and in 2011 they completed a five-day Quick Reaction Assessment.

In December 2011 an unmanned K-Max was reported to be at work in Afghanistan. [...]

On July 31, 2012, Lockheed announced a second service extension for the K-MAX in Afghanistan for the Marines. This extends operation time to the end of March 2013, with the option to extend through to the end of September 2013. As of July 31, 2012, Lockheed reports the K-MAX has completed 485 sorties, 525 flight hours, and delivered over 1.6 million pounds of cargo since its deployment in November 2011"

The Army uses tons of transport helicopters in non-combat situations though, right?

The military (having a significantly lower standard for 'safe enough' operation than, say, airlines) will be an interesting inroad for autonomous transport vehicles in society.

Sure, and there may be applications there, but they aren't testing with a transport helicopter (such as a Chinook which flies quite differently than a Blackhawk or other traditional helo design), they're testing Blackhawks in a terrain avoidance environments. Two wholly different things.

What's important here is how you're using and what encapsulates "transport", that imo is critical to the point you're wanting to make, and whether or not I disagree with it.

The military is usually the inroad for wide use civilian technologies.

Either you're saying most civilian technologies were military in origin. Or you're saying most military technologies end up with civilian applications.

I doubt the latter is true, and I'm not sure about the former either. The military R+D budget is big, but these days (maybe always?) it pales compared to public nonmilitary and commercial R+D activity.

I don't know, I think the second one could be true. I imagine it's quite likely that someone involved with this is also a hobby drone builder that also contributes to open source. If they've created some new algorithms or made some existing ones more robust it would be great to get those to the community.

Your right, and I think where this will end up is with 'super aware' pilots. Where the plane does just about everything automatically, but there's a human inside that basically 'thinks' and the vehicle does. Our brains are already better at strategy than the most advanced computers, it makes sense that the vehicle takes care of the easy stuff (like not hitting walls and landing) and the human focuses on the enemy.

"Our brains are already better at strategy than the most advanced computers"

--- for how much longer? and when taken at scale, is the human race really all that great at strategy? We are so much at the mercy of emotions to ever be truly great strategists. What if, for example, given food production and water usage etc the math dictated that the population must reduce, do you really think this strategy would be adopted? Of course not.

i don't really see the point of self-flying helicopters, it's not like there's a shortage of bright young men and women who want to fly these things. The real challenge isn't flying helicopters, but finding ways to resolve situations without blowing shit up. As fun as that might be...

> --- for how much longer? and when taken at scale, is the human race really all that great at strategy?

We still mop the floor with Go programs and AI in RTS games.

When the hoard plays nasty in Left for Dead, I can't say I can beat AI.

In most games, the deck is stacked in favor of AI in a variety of ways. (BTW, it's "horde," unless it's a deliberate use of a homophone by the game designers.) If the AI's units can overwhelm humans with superior reaction times or superior numbers, then it's not the AI's strategy that's winning. There are real world situations and games that are still complex enough, it's difficult to compute tactics and strategy properly.

Here's the difference between humans and AI in a nutshell: Competent humans can usually tell when someone is using an exploit against them. AIs almost never have a clue that's happening.

2 good points in there, and as a zombie fan, I'm ashamed of my horde failure. And the comment on exploits is sharp, has AI ever noticed exploits/cheating? How? I know how I start to suspect and start to confirm suspicions, but its not much more than recognition of suspicious patterns.

"I don't really see the point of self-flying helicopters, it's not like there's a shortage of bright young men and women who want to fly these things."

Fewer caskets arriving at Dover AFB for repatriation.

While that may be an issue at higher political levels (deaths make wars less popular), young people aren't put off by the risk of death, as armies have know for thousands of years. And given the description of the US military as being Americas social welfare system, is this for the greater good, or not?

How about where troops are pinned down in an extremely hot firefight and running low on ammo. The commanders aren't going to risk letting a human piloted black hawk land because it's just too damn risky, but a drone might get the go ahead.

It also makes sense from a purely rational point of view because the major costs of that capability are pilot training (millions) and maintenance. It doesn't hurt nearly as much to lose a chopper when you don't lose those millions spent in pilot training along with it, so you can take much greater risks with it

I agree that this will (and should) end up with 'super aware' pilots. There is a lot of work that can be and should be automated so that a pilot only does what is absolutely necessary -- tactical strategy and anticipation of human responses. Computers and their programming can be made mathematically perfect. Flying, evading, tracking, and even self-preservation are perfect for this.

The one thing that a human can do is make real mistakes, anticipate such errors, and make seemingly illogical choices -- illogical to a computer. True deceit. A computer may be able to one day fake deceit, but there will be a logical probability factor to it. Detection of such deceit will likely be more difficult for a computer algorithm than faking deceit.

Vaguely reminds me of Snake Eyes, by Tom Maddox, though not quite as extreme.

http://www.dthomasmaddox.com/snakeeyes.htm

Where the plane does just about everything automatically, but there's a human inside that basically 'thinks' and the vehicle does.

"...must think russian"

I think initially this could be more effectively used for transporting the helicopters from one place to another, such as another base.

You are right though, aircraft like the Global Hawk operate at altitudes that are safe from enemies, so operationally they are easier to reason with for autonomous use.

Reply on news.ycombinator.com