Slashdot Mirror
Cornell Builds Autonomous UAV
tshak writes "From Microsoft Research, 'Faculty and students at Cornell University have built an unmanned airplane with its own on-board, embedded control system. The large-scale model plane flies by accessing coordinates from an off-the-shelf GPS unit.' Not only does the plane run XP embedded, but the software is written in C# on the .NET Compact Framework. This is all powered by an 800mhz Crusoe processor with 1GB of total system storage."
http://avdil.gtri.gatech.edu/AUVS/IARCLaunchPoint. html
"This year the best performances were executed by the Georgia Institute of Technology and the University of Arizona (first year in the Competition). The Georgia Institute of Technology's autonomous helicopter demonstrated "Level 2 behavior" (finding a particular building from among many and then identifying all of the real openings in the building through which they could send in a sensor probe) during a series of three flights comprising more than an hour of flawless fully autonomous operation-- they landed only twice, once to adjust a camera and once to top off their fuel tank."
My first post (yay! - now modded 0 - boo!) was only intended to emphasise what the linked site acknowledges:
"Safety - An autonomous flight control system inherently removes human operator intervention from vehicle functionality. The capability of an aircraft of considerable mass, traveling at high velocity, to inflict damage to people or property is substantial. It was critical that the flight control system would include several modes of flight termination in case of emergency or flight control system failure."
Sure, there are non-military uses for UAVs - crop dusting, fire fighting, ground mapping, etc, but I'd be interested to know what proportion of the devices currently out there are in fact military.
Phil
Obviously the Microsoft Research link gives away the "why", but I really want to know why the overhead of .NET? Why not just compile it to "real" C++, build some machine code, and stick it on a chip? Doesn't that make more sense than .NET? Plus, I'd like to toss in my $.02 that 1GB is totally insane for an embedded system. The space shuttle doesn't need that much for its automated stuff.
stuff |
I'm sure they could start from the codebase they have now, work really hard, and have equivalent expertise built into a lighter package in some number of staff-years. Alternately, they could archive the source, go drink margaritas for a couple years, and then buy the lighter package with equivalent power off the shelf and use it to run the existing already-completed software.
Which would you consider the elegant way to proceed?
"Consider yourself a member of a virtual corporation with Mr. Torvalds as your Chief Executive Officer." - Linux Advocac
Mission critical how? I've never written a bomb tracking program in C#, but I've never been asked to. With that said, I'd trust .NET as much as I trust Java or Perl -- the framework is quite robust and the execution environment (CLI) is reliable.
.NET that have been up on client machines for months. Garbage collection means no danger of memory leaks (and no, I've never seen a slowdown due to GC, it is much more subtle than GC in Java 1.3 which would basically shut down your app for a few seconds), and all of my socket manipulations are wrapped in self healing stubs that catch failures and exceptions and restart the information queue where it left off. Our clients have never seen a bug in the data layer or the framework itself (though our presentation layer has had plenty of bugs...we follow the Microsoft development philosophy here, get it to clients on day X with an acceptable feature-to-bug ratio so they can start getting the most essential work done and we can fix the rest incrementally).
I have programs I've written in
This is using Framework 1.0; there have been two versions since then with performance, stability and feature enhancements.
Hey freaks: now you're ju
I think it's be cause a lot of people (not all) here see it as a waste. The time/effort spent making it run XP embedded, and making the hardware accomidate all that storage space, could have spent making it run a light OS. XP Embedded is small, but last I heard bare-minimum was still several MB. A lighter OS could have fit on smaller hardware, and provided more space for other stuff, or a small plane design in all.
.net implementation there are several linux ones available as well - or Java could have even been used. A lighter CPU could have been used (or several), and it would have been a much more impressive feat.
.NET runtime isn't a well-enginieered solution. It _IS_ however impressive that consumer-level products can be used for such a thing. Interesting in it's own way. And I'm sure it had it's own hurdles. It's just that from the enginieering point of view, the initial reaction is "Well, yeah, but what a waste".
Even if they wanted to use a
Throwing XP embedded, a 1 gb drive, 800 mhz CPU, and
You're right. 1G of storage - that's equivalent to two whole USB keychains. It couldn't possibly get off the ground.
Here's how I see it. It's a freakin' prototype. It doesn't have to be elegant. It just has to work. And maybe they want to add other functionality later, such as "look out for that tree" and "don't fly into the building, even if it's a waypoint".
Besides, autonomous flying machines need all that extra power to run Skynet anyway.
144l. ph34r my 133t l3g4l 5k1lz!
My school(Virginia Commonwealth University) also did this, and actually won the competition these were built for, the second AUVSI student competition./ default.htm/ uav.html
http://auvsi-seafarer.org/seafarers
http://www.egr.vcu.edu/announcements
As far as I know, no Microsoft products were used on the plane, but I can't find too many details at the moment. The guy I know who worked on the project only knew C and C++, though from my understanding he did mostly the EE stuff, not as much programming....
Blessed are the pessimists, for they have made backups.
Also, who uses some grossly huge OS for real-time embedded applications? Typically there isn't even an OS, or people use very small things like OSEK, which are basically just interrupt handlers that schedule all the tasks you need.
I just don't think people know how to program any more, when they can't even figure out how to write solid code and avoid buffer overruns and decide to use a "sandboxed" language to make up for their (laziness? lack of competence? management directives to work so fast time for care is squandered?).
"There are a dozen opinions on a matter until you know the truth. Then there is only one." - CS Lewis (paraprhase)
Credit where credit's due - it's a truly great feat of engineering.