Robot Planes and Helicopters Taught Aerobatics

holy_calamity writes "MIT and Georgia Tech researchers are teaching small robotic aircraft some impressive stunts. MIT's RC plane's can take off and land from vertical perches (video), while the Georgia Tech helicopter can land on slopes of up to sixty degrees, by flipping backwards into freefall as it lands (video)."

Comments

One of the best Helicoptor pilots

[Anonymous Coward]

http://www.youtube.com/watch?v=p8t41avFuCc

  Alan Szabo Jr

Re:One of the best Helicoptor pilots

[chuckymonkey]

That was amazing. The military needs to figure out how he does all of that. While a human would not be able to take the forces something acting like that would exert, a remotely controlled fighting craft would be incredible. Back on topic that's some pretty interesting stuff. Most UAV require a landing strip like an airplane to take off and land on, if they could just do what these do it would make them much more portable and much easier to use in the field. Amazing world that we live in these days. My dad couldn't imagine having a phone in his car when he was a kid in the 50's much less aircraft that could fly themselves around.

Re:One of the best Helicoptor pilots

[MouseR]

Woah. Just, woah.

Re:One of the best Helicoptor pilots

[stoolpigeon]

Some of the UAVs that you see in the media a lot like the predator and global hawk do need some space to operate, due to their scale. But I don't think it is safe to say that most UAVs are that large or require a runway. There is a whole range of Miniature UAVs that can do some slick stuff. I guarantee if you're seeing it on youtube and thinking of possible military applications, the military has probably already made that connection some time ago. There are a lot of people, some of them pretty smart, who have dedicated their lives to this kind fo thing.

Re:One of the best Helicoptor pilots

[jcr]

UAVs have to make a trade-off of size for endurance. A UAV like the predator can stay aloft for hours, but something small enough to hand-launch would be impressive if it could even carry enough fuel to stay aloft for half an hour.

-jcr

Re:One of the best Helicoptor pilots

[superwiz]

Why is Wagner playing in my head while I watch this?

Re:One of the best Helicoptor pilots

[jcr]

*I* have built R/C planes of the hand launched variety (3-5 foot wingspan) that can stay aloft for 2+ hours on *BATTERIES*

That's nice. What was their on-board avionics complement like? How many cameras, and did you have visible light only, or IR and UV sensors as well? How about chemical sniffers, GPS, or encrypted and jamming-hardened radios? How about motion compensation? Autopilot if control link was lost?

What you can build for hobby purposes, and what's practical for a military mission are very different things. Your R/C planes only have to have a receiver good for visual range (a couple hundred meters at most), and you don't even have any downlink.

I've built R/C planes too, and I also worked on the ground control station for a target drone UAV many years ago. You're the one who doesn't know what he's talking about, son.

-jcr

Re:One of the best Helicoptor pilots

[Sparr0]

onboard avionics consisted of just a three-axis gyro based autopilot (in case of control loss). cameras were medium-resolution still and low-resolution radio-transmitted video, visible light only (well, as little IR as I could manage, you know how CCDs are). gps was only for tagging with the camera, a small non-interactive reciever.

oh yeah, and i flew well outside visual range. thats what the video camera is for. radio and reciever were tested to 1.5 miles, probably good to at least 2 on a good day.

the army can have as many as they want for $2000 apiece. screw hardening, the enemy can knock down 99% of them and they will still be cheaper than any "real" military UAV ive heard of.

Re:One of the best Helicoptor pilots

[Cederic]

For tactical use, something small enough and cheap enough to have in a box in the back of the truck for immediate recce purposes might be of interest. A platoon commander would love to be able to answer "What's happening half a mile over there" at will.

Of course, I'm mostly just guessing.

Re:One of the best Helicoptor pilots

[Sparr0]

About as well as a proposal to equip soldiers with silly string would go. Laughed off out of hand, impossible to accomplish officially, but wildly useful when actually delivered at private expense.

Screw the $2000 version. For $250 (each, in quantities of 20+) I can give you a UAV with a 30 minute flight time and remote video at a 1 mile range that will fit [no-tools disassembled] inside a shoebox for carrying*, with a MTBF of 1000 hours [in the air]. $10 each for sets of rx/tx crystals, leaving channel allocation as an exercise for the reader. $100 more gets you the transmitter (the radio control), a battery charger, and a video monitor, all of which fits in a second shoebox, but each unit only ever needs one of those (assuming the UAV itself is going to get lost on a regular basis). As a sibling post stated, I don't think any ground unit would not want to have that as a tactical option, and we spend more than that on a good periscope just for looking around a corner. "milspec" is fine and dandy for some things, but we suffer in wars like Iraq and Vietnam where the enemy can do more with $10 than we are willing to do with $1000.

* - given a 14"x8"x6" shoebox... With a 3 segment foam wing that buys you you a 42"x8" wing, a 2 segment boom gives you at least a 32" long fuselage, and the rest of the box is taken up by a two-segment tail, the electronics and mechanical packages. It would carry 1Ah*11.1V in lithium batteries (oh no, a fire hazard!), run on a 5-7" prop geared down maybe 4:1 from a brushless motor. Control would be throttle+pitch+yaw, with a respectable amount of dihedral and COL/COG separation to avoid roll.

PS: Could almost certainly do significantly better on the range with frequencies available to the military. 1 mile is easy with the public no-license-required bands here in the states, and trivial on ham R/C bands.

PPS: Jamming will be a problem. Add $50 to the non-replaceable package cost to get a transmitter that can operate in multiple bands, have different planes for different bands (no extra cost). If XXXMHz is being jammed, fly something on YYYYMHz instead. If they manage to jam every available frequency... you're no worse off than you started. And you can probably get some sort of useful information out of the location of the jamming equipment (trivially triangulated).

Not RC

[SnoopJeDi]

I see nothing that suggests that the MIT plane is remote-controlled. It was inspired by a pilot's skill on an RC model.

Unless the controls are issued by a remote computer?

Re:Not RC

[The Raven]

It is absolutely remote controlled. If you RTFA, it states:

testing indoors with off-board control systems and sensors. "The bit in the air is the cheapest part of these experiments," he says.

The controls, cameras, everything is not on the plane. The plane/heli are just simple cheap RC toys, controlled remotely by expensive processors and sensors.

I, for one, ...

[Anonymous Coward]

...welcome our small robotic aircraft overlords!

But, I suspect that we'll soon be chased around by flying advertisements!

[mechanical voice]: "Wait, Mr. Smith, stop running! I've got to tell you about Splam!"

(Sound of one flying ad machine shooting down another)

[second mechanical voice, swooping in]: "Don't listen to that guy! Splastic is the new Splam!!!"

milestone

[giampy]

The videos are very interesting, especially the second one from the group of Jonathan How. The developmen of control laws that are able to fully control the aircraft flying in those conditions, (not to mention being able to handle the transitions between such flying modes) is a hard problem.

This is due to the fact that the overall system is highly nonlinear, scarcely controllable, (since the control surfaces have little to no effect), and also not very well known in such conditions.

Whenever they can handle this problem in a systematic and rigorous way, (that is without ad-hoc quick fixes), i'd say that a milestone in control science will have been reached.

Re:milestone

[timeOday]

Whenever they can handle this problem in a systematic and rigorous way, (that is without ad-hoc quick fixes), i'd say that a milestone in control science will have been reached.

The opposite may be true; I think it most likely this was achieved by machine learning rather than control theory using programming by demonstration, reinforcement learning, etc. Classical linear control theory is nice and formal, but formalisms can limiting rather than empowering when you fail to think beyond the narrow limits of what you are able to prove. In other words, in practice you might get a whole lot further by giving up on guarantees (given assumptions which are almost certainly unrealistic anyways); and using more "ad-hoc" methods.

Re:milestone

[giampy]

I am not sure i'd call that AI. A number of animals like for example insects exhibit an amazing control of their body within a wide variety of situations, but "intelligence" in the usual sense does not have to be involved. I'd call it simply sensing, motion planning and control. By that i don't mean is a simpler problem than AI, (it may very well be harder), i just think it's different.

Autonomous learning techniques can help, to a certain extent, but i personally think that they cannot do the whole job, as control theory cannot be completely bypassed. Animal brains have already in place dedicated motion planning structures, evolved trough billions of years, that allow them to do what they do. THen it takes a huge amount of trial and errors to perfect their "control system". However this is an open issue. We'll see.

By scarcely controllable i meant that when the airplane is hovering like an helicopter its control surfaces (ailerons, stabilator, rudders) do not receive considerable air flow, and therefore you cannot really expect a substantial "return force" on the airplane by moving them. Witout these return forces, you (or whoever or whatever is trying to control the airplane) are very limited in moving the airplane around.

By the way, the other poster (slashdot.org) was perfectly right, sensing (especially indoor without gps) is another huge challenge here. Machine vision (using the leds but eventually even using features extracted from pictures of the environment) can help.

However, even with perfect magical sensing, the motion planning, guidance and control problems cannot be handled by "classical" control and require techniques that are currently on the very leading edge of modern control theory (robust, optimal control of nonlinear and hybrid systems and the like).

G.

Can I get that software...

[kclittle]

... in the flying car I made a $10K deposit on?

Convair Pogo

[plsuh]

Just for historical reference, the Navy experimented with something like this back in the 1950's. According to the writeup from the Smithsonian, the Pogo suffered from a lot of control problems due to propwash buffetting near the ground at takeoff and landing. Back then it took a very skilled test pilot to keep it under control; modern flight control systems like those used to keep semi-unstable airframes (such as the F-16) in controlled flight must make similar VTOL handling a lot easier today.

http://www.nasm.si.edu/research/aero/aircraft/convair_pogo.htm

--Paul

20 degrees?

[v1]

The second video says the best real helicopter can only land to 20 degrees. There are two things I have to ask about that and they both relate to why it works with the model.

(1) how much of that limit has to do with the design of the model helis? I fly model helicopters, and they (tho not I !) are capable of inverted flight by pitching the main blade the other way. I assume the model in this video is using this method to drive the heli down and pressing it against the slope when it has touched down. It's quite possible there's a pressure switch on the bottom of the skids that jacks the pitch the other way when it makes contact with a surface. Not a bad idea really, and this change happens in a VERY short period of time. You don't see any full scale helis capable of inverted flight, no doubt due to the mechanical difficulty in making the main rotor able to support the weight of the craft in the inverted position. The fact that the video does not show the heli taking back off again makes me seriously wonder if there isn't a contact switch at work.

(2) kinda dark in that video, I wonder what sort of surface they were landing on? Surely not velcro. Maybe a rubber mat? Probably a lot easier to do that, especially with a light craft, than on say a steep grassy slope or dirt hill. And what was on the bottom of the skids?

I'd be interested to see some statistics on the power-to-weight-ratio and such comparisons between a model heli and a passenger heli also.