Crash-​​Proof UAVs Fly Blind at MIT

Here’s another great story from our friends at Popular Mechanics that looks at cut­ting edge research into drones that fly autonomously inside struc­tures. That’s some­thing that until now could only be done (barely) by wheeled mini-​​bots. But as you can see from this report, engi­neers still have a long way to go.

It’s not the most attrac­tive spy bot, but the unmanned aer­ial vehi­cle hov­er­ing some 20 ft. away is doing its job. For now, that means stay­ing right where it is, weav­ing ever so slightly under the weight of the web­cam strapped to its back. There’s noth­ing par­tic­u­larly inter­est­ing to look at with this UAV, a com­mer­cial four-​​rotor model that any RC hob­by­ist could put together. But no one is pilot­ing this mod­i­fied drone — it’s fly­ing autonomously, sta­bi­lized a few feet above the floor of MIT’s RAVEN lab. Like most of the air­craft tested here, this model is a pup­pet, receiv­ing input not from onboard proces­sors, but from a nearby com­puter.

As it con­tin­ues to buzz in place, an array of 18 motion-​​capture cam­eras tracks the UAV, pro­vid­ing 3D posi­tion­ing data to deter­mine just how sta­ble it is. Specifically, those bale­ful red cam­eras — the same kind Hollywood visual effects teams use to trans­pose an actor’s move­ments to a computer-​​generated coun­ter­part — are track­ing the tiny Styrofoam balls attached to the drone. On the com­puter mon­i­tor, these balls show up in real time, map­ping the UAV as a clus­ter of dots, sway­ing in midair. I’m some­where between impressed and bored when the drone begins to drift. A sec­ond later and it slams into a plex­i­glass divider, as hard as a hockey player.

It will take some time to fig­ure out why this lit­tle craft sud­denly lost con­trol. But that’s the point of RAVEN, or Real-​​Time Indoor Autonomous Vehicle Test Environment, where geeks cap­ture every flight — and col­li­sion — in painstak­ing detail. There are no acci­dents here, just prob­lems that haven’t been suf­fi­ciently ana­lyzed. “RAVEN gives us the free­dom to test what­ever we can build,” says Jonathan How, direc­tor of MIT’s Aerospace Controls Lab. “And we can build won­der­ful things, even in 24 hours.“

One of the researchers has done just that, and is now prepar­ing to fly a drone that was redesigned, then cob­bled together out of light­weight foam core. Of course, this isn’t exactly the next gen­er­a­tion of missile-​​packing Predators; the toy­like cre­ation in front of me, with its cir­cu­lar wing and minia­ture nose-​​mounted pro­peller, is more of a test­bed than a pro­to­type. All of the UAVs cov­er­ing nearly every sur­face of this lab, from high-​​end RC planes the size of a small child to a store-​​bought fly­ing insect pro­duced by WowWee, are just tools to develop flight con­trol algo­rithms for indoor robots.

As chal­leng­ing as it is to make some­thing fly itself, design­ing a drone that can func­tion indoors is even harder. For an indoor UAV to meet all of the military’s expec­ta­tions, it would need to be able to fly into a build­ing and find a suit­able spot to perch and observe, all with­out rely­ing on GPS con­tact. “The ulti­mate vehi­cle is a bat that you can down­load data from,” How says. Bats have the abil­ity to perch, plus echo loca­tion to detect obsta­cles, and the agility to keep from slam­ming into them. 

At the moment, noth­ing in devel­op­ment can effec­tively pull off even one of these func­tions, much less all three. Researchers at Carnegie Mellon University, for exam­ple, are stick­ing to basic nav­i­ga­tion, with a small robot heli­copter that uses sonar and cam­eras to avoid bump­ing into obsta­cles while fly­ing indoors. But the lab here has a novel approach: Instead of focus­ing on build­ing bet­ter sen­sors or more pow­er­ful vehicle-​​mounted proces­sors, researchers at RAVEN are fine-​​tuning the mechan­ics of autonomous aer­o­bat­ics. The 18 motion-​​capture cam­eras pro­vide a per­fect sens­ing envi­ron­ment, and the ded­i­cated com­put­ers in the lab, which com­mu­ni­cate with the test drones via radio trans­mit­ters, pro­vide the brain­power. “People might say our UAVs aren’t autonomous,” How says, “but in this envi­ron­ment, the entire sys­tem is autonomous.”… 

Read more on this story and other high-​​tech reports from our friends at Popular Mechanics.

– Christian

April 25th, 2008 | Popular Mechanics | 281026 Comments »http://defensetech.org/2008/04/25/crash-proof-uavs-fly-blind-at-mit/Crash-Proof+UAVs+Fly+Blind+at+MIT2008-04-25+14%3A07%3A50Ward You can skip to the end and leave a response. Pinging is currently not allowed.