Slashdot Mirror
DARPA's Atlas Walking Over Randomness
mikejuk writes "Considering how long we have been trying to solve the problem, a robot walking is mostly amusing. Atlas is an impressive robot, evoking the deepest fears of sci fi. Watch as one of the DARPA challenge teams makes Atlas walk, unaided, on randomness. This video of Atlas was created by the Florida Institute For Human and Machine Cognition robotics team. It shows Atlas walking across a random collection of obstacles. Notice that even though it looks as if Atlas is supported by a tether, it isn't — as proved when it falls over at the end."
Those flat board feet are no match for the terrain. A few metatarsals would go a long ways toward stabilizing its stance on uneven surfaces.
The determined Real Programmer can write Fortran programs in any language.
It's understood perfectly. It's just very very hard to simulate, as there are many muscle groups involved from your toes up to your head and your shoulders, all moving in a coordinated effort. Getting it exactly right on a robot is - hard.
Seven puppies were harmed during the making of this post.
An impressive balancing act, but the really interesting thing is to see how Atlas stumbles on objects which shift under its weight. It vividy shows the need for flexible feet on a biped. That's going to be an interesting engineering puzzle to solve.
The most recent Big Dog video shows that this is a (largely) solved issue for quadrupeds. The middle of the video shows it walking over stumps, navigating a swingset, and so on.
Still, "four legs good, two legs better". Or so they say...
The summary is an impressive bit of (unattributed) quoting from the article, evoking, the deepest fears of comma abuse.
hardly...
It might freak the cat out in a similar vein to a vaccuum cleaner but wathcing the video makes me feel worried *for* (not of) Atlas
I'm sure it's an amazing technical achievement but just highlights how much further we have yet to go.
One thing I haven't seen on either this or ASIMO is a back with lordosis. This is critical for our balance while walking, and may allow these robots to get rid of the knee bend.
Shouldn't the random obstacles include human skulls?
We do have a shortage of engineers. There's a distinct shortage of engineers willing to work unlimited hours for laughably low wages under demoralizing petty authoritarian management, after spending years of their life and racking up $100+K in debt on the promise of decent employment. This shortfall is made up for by the H1B program.
also, pretty much every human being would make sure to not step on any of those obstacles, mainly because they don't want to trip and fall. Who seriously sees a corridor full of wood and goes "fuck it, i'll just stroll straight over all this"? You don't, you look to make sure you're stepping on solid ground. Sure, there's times when you do have to tread on uneven and unsecured surfaces, but then you take even more care on where you're going. This robot doesn't seem to have the logic to do this, and that's where they're going wrong.
If you gave me a choice between a printer and a giraffe with explosive diarrhoea, i'll get my ladder and my raincoat
Only if reductionism works for walking. There are some systems where simplifying the system removes the interesting behavior. You may be able to understand the complex system, but be unable to build a replica due to physical limitations.
Not a sentence!
Notice that even though it looks as if Atlas is supported by a tether, it isn't - as proved when it falls over at the end.
...and then proceeds to hang by its tether.
... with its center of mass at a lower altitude than when it was walking, thus proving that the tether wasn't supporting* it before, yes?
*I'm aware that proves it wasn't entirely supporting it, and there'd need to be some assumptions made about how elastic the tether is to prove it wasn't supporting it at all, but given that the point of such a tether is to arrest a fall before robot meets concrete, and also to not impose any force before the fall, I think it's reasonable to suppose it actually does that.
That's the amateur version from the Florida Institute For Human and Machine Cognition. Here's the pro version from Boston Dynamics, showing some walking over rocks. The balance control is better, the walking is faster, and the arms and torso are being used more effectively for balance.
The "DARPA Humanoid Challenge" teams are struggling along. They had to write their software to run in a poor simulator, then use it on the real robot, with a competition next month in December. So the control software is crude. Most of the team efforts seem to be going into the perception side. Performance in the simulated humanoid challenge was poor; the best team fell down about 12 times. This looks like they're still using the basic balance controller from Boston Dynamics for control. Entrants in the competition get a closed-source .so file that will operate the Atlas robot for a few basic functions (slow walk, stand, etc.) for debugging purposes. This isn't the good stuff; Boston Dynamics keeps the better algorithms a secret. Entrants are supposed to replace those algorithms with better ones, but since they've only had about two months with the real robot, that probably hasn't happened.
In a year, this will probably suck a lot less.
One day, robots will be ubiquitous: everywhere, all the time, helping in whatever ways are necessary. (Yes, I did read a lot of Isaac Asimov growing up. But it still seems inevitable.) Bipedal robots will be necessary, since our artificial world has been constructed for bipedal creatures between about 3' and 7' tall.
Japan's Asimo robot was a ridiculous dead-end: perfectly controlled to essentially run scripts over well-known terrain with virtually no uncertainty or randomness. A real bipedal walking robot will need uncertainty baked in from the very beginning, which Atlas clearly has from watching the video. It looks like a (slowed down, more clumsy) version of a person walking over uncertain terrain, if that person were blindfolded (which Atlas essential is in this demo, as it has no camera inputs). What you see as evidence of weakness is actually evidence of vast ability!
From these beginnings, you're going to see researchers add in more complex terrain coupled with video cameras input for prediction of foot-ground-interactions, and gradually increasing speed. Around the end of this decade (or perhaps the next), we will see bipedal robots able to work and move (including running) in and over highly variable natural terrain, and thus be truly useful in whatever situations they are required, thanks to the productive research direction pushed forward by Atlas. Yes, it will take some time, but this is one branch of artificial intelligence / machine learning, and as we have discovered in the field, the things that humans and living things find easy are actually incredibly *hard* to get robots / computers to do.
No human has vibrating feet.
I beg to differ. Try this experiment. Place a thick pillow on the ground, or perhaps two. Stand on these pillows with one leg and no other support for one minute. If your pillow stance is unstable enough, your foot will have to move around rapidly to maintain your balance, since by not being able to change the location of your foot on the pillow, you must instead change the orientation of your foot. I suspect that this robot is actually testing specifically the ability of ankle joints to maintain balance, since there is almost no side stepping visible on the part of the robot. The Boston Dynamics robots referred to in other posts often seem to rely mainly on sidestepping for balance, and often have peg legs instead of feet and ankles. This robot seems likely to be a proof of concept of one particular method of balance that in future robots will be combined with other methods of balance.
This and no other is the root from which a tyrant springs; when first he appears as a protector - Plato (423 to 327 BC)
The first attempt at DARPA Grand Challenge autonomous car race (http://en.wikipedia.org/wiki/DARPA_Grand_Challenge) made it less than 12km before getting stuck - in 2004. Now only nine years later people are talking about the imminent arrival of driverless vehicles.
I've made that point before. I was at the 2004 DRC (and in the 2005 one). The 2004 DRC was pathetic. It was covered by the Comedy Channel. The Ohio State entry (a huge Oskosh truck) ran into a parked vehicle at slow speed and pushed it for a while until DARPA people finally sent it an emergency stop signal. The CMU approach was to have a semitrailer full of people at workstations doing detailed manual path planning. The CD with the route was released an hour or so before the start, so their people had a short period to plan the exact path, using recent high-res aerial photos. DARPA's competition chief, an active-duty USMC colonel, found out they were pre-planning, and so, the night before, a few of his troops went out and moved some obstacles. This was the result. CMU's vehicle plowed right through a highly visible sheet metal fence. They were the most successful team. The others did much worse.
Then in 2005, there were 23 teams with working vehicles running around the California Motor Speedway, none running into anything. The second day of the 2005 Grand Challenge was the day the press suddenly recognized that automatic driving was real.