Smarter Robot Arms

← Back to Stories (view on slashdot.org)

Posted by Soulskill on Friday September 23, 2011 @03:36AM from the rock-em-sock-em dept.

RogerRoast sends this quote from the MIT News Office: "As intuitive as it seems to a human being, spontaneously planning a trajectory around obstacles in free space is a monstrously complex computation. As a consequence, most motion-planning algorithms give up on the idea of finding the most efficient path between the robot’s initial state and its goal, settling for any path that won’t introduce collisions. [Researchers at MIT] have built a new robotic motion-planning system that calculates much more efficient trajectories through free space. ... Not only do robots guided by the system move more efficiently, saving time and energy, but they also move more predictably (PDF), a crucial consideration if they're to interact with humans."

52 comments

Min score:

Reason:

Sort:

Kids aren't that good at it by Mr.+McGibby · 2011-09-23 03:49 · Score: 1

I have small children (up to 9) who don't seem very good at path finding. They're constantly going around the house in the most inefficient manner possible. Humans aren't that good at it either.

--
Mad Software: Rantings on Developing So
1. Re:Kids aren't that good at it by instagib · 2011-09-23 03:53 · Score: 0
  
  OK, let me understand this: you have children, not sure how many but max. 9, and they all circle your house, not finding the door? Fascinating! ;)
2. Re:Kids aren't that good at it by Anonymous Coward · 2011-09-23 03:57 · Score: 1
  
  can't blame you since the summary is a bit misleading, but this is not really about "path finding" in the sense of navigating a body around an environment, it's about efficient movement of a manipulator at the end of a robot arm to get to it's target. Every joint that introduces a degree of freedom (this robot had 7 DOF arms) makes that calculation more difficult.
  Closest analog I can think of is something like a back hoe operator. The difference between a good one and someone just learning is huge in terms of how smoothly and precisely they can move the digger in space.
3. Re:Kids aren't that good at it by Mr.+McGibby · 2011-09-23 04:06 · Score: 1
  
  Fascinating enough for you to comment.
  
  --
  Mad Software: Rantings on Developing So
4. Re:Kids aren't that good at it by Mr.+McGibby · 2011-09-23 04:07 · Score: 0
  
  That was creative.
  
  --
  Mad Software: Rantings on Developing So
5. Re:Kids aren't that good at it by Anonymous Coward · 2011-09-23 04:07 · Score: 0
  
  Or ADHD, or too much sugar/crack. Probably should give them some downers or xanax, or beat some sense into them.
6. Re:Kids aren't that good at it by NoNonAlphaCharsHere · 2011-09-23 04:10 · Score: 1
  
  I was at an industrial robotics trade show years ago, and there was one spray-painting an auto bumper that amazed me how gracefully it moved; at the end of a swing left-to-right, it turned its wrist at the same time it reversed the arm direction. The only word I could apply was "graceful". Later, I found out it had been trained by a human operator, so it was just mimicking what a human spray-painter with years of practice would do.
7. Re:Kids aren't that good at it by Anonymous Coward · 2011-09-23 04:17 · Score: 0
  
  Keep in mind, the robotic spray painter works in a well-defined environment without pesky people walking around getting in the way. It only has to do one task over and over. In this scenario (the bulk of industrial robotics), the path planning is static. You do plan it once for a task and then it just loops. You'll only need to change it if you move the painter to an assembly line for another vehicle.
  A vastly more complicated process is to handle dynamic path-planning. In this scenario, the environment and the task change regularly.
8. Re:Kids aren't that good at it by mjwx · 2011-09-23 04:34 · Score: 1
  
  OK, let me understand this: you have children, not sure how many but max. 9, and they all circle your house, not finding the door? Fascinating! ;)
  Clearly one has never lived with children sub age 10. They can be relied upon to find coffee tables, sweets and their sisters hair but that's about it.
  
  --
  Calling someone a "hater" only means you can not rationally rebut their argument.
9. Re:Kids aren't that good at it by pnewhook · 2011-09-23 04:39 · Score: 0
  
  I thought your comment was pretty funny. Unfortunately you had a bunch of complete shitheads (especially AC) responding.
  
  --
  Tesla was a genius. Edison however was a overrated hack who liked to torture puppies.
10. Re:Kids aren't that good at it by Omegawar · 2011-09-23 04:40 · Score: 1
  
  9 isn't the max. They could have had twins, triplets, or even more. But then again 10+ pretty much guarantees you a reality show.
11. Re:Kids aren't that good at it by fahrbot-bot · 2011-09-23 04:42 · Score: 1
  
  Your children are in Roomba mode. Find the instruction manual and reset them.
  
  --
  It must have been something you assimilated. . . .
12. Re:Kids aren't that good at it by Psion · 2011-09-23 05:04 · Score: 0
  
  I'd listen to this guy, McGibby. If anyone knows stupidity, it's him.
13. Re:Kids aren't that good at it by jeffmeden · 2011-09-23 05:29 · Score: 1
  
  "As intuitive as it seems to a human being, spontaneously planning a trajectory around obstacles in free space is a monstrously complex computation"
  I was going to say a similar thing; observe any major highway in any major city in the USA and you will realize that many many people are absolutely terrible at "planning a trajectory around obstacles in free space"... So bad that it causes as many as 90 deaths *every day*.
14. Re:Kids aren't that good at it by hitmark · 2011-09-23 05:54 · Score: 1
  
  Not helping that our sensory processing is not really built for anything beyond a walk. Running, much less driving a vehicle at highway speed, can't be properly processed beyond a straight and flat path.
  
  --
  comment first, facts later. http://chem.tufts.edu/AnswersInScience/RelativityofWrong.htm
15. Re:Kids aren't that good at it by Anonymous Coward · 2011-09-23 06:04 · Score: 0
  
  Don't forget the cat food. They can find the cat food and water pretty reliably. And bits of fuzz.
16. Re:Kids aren't that good at it by JackCroww · 2011-09-23 06:15 · Score: 1
  
  9 = max age, not max quantity
  
  --
  "Ayn Rand is a bloody socialist compared to me." - Robert A. Heinlein
17. Re:Kids aren't that good at it by bugs2squash · 2011-09-23 06:42 · Score: 1
  
  You think they are being inefficient because you don't understand the constraints; in particular any carpeted are of floor is "red hot lava" and sofas are bouncy, allowing "slingshot" style trajectories that would be the envied by NASA.
  If your children are not human, what does that make you ?
  
  --
  Nullius in verba
18. Re:Kids aren't that good at it by sgt+scrub · 2011-09-23 06:44 · Score: 1
  
  Kids that need rebooting? I sense a patent lawsuit in his future.
  
  --
  Having to work for a living is the root of all evil.
19. Re:Kids aren't that good at it by Lunix+Nutcase · 2011-09-23 07:11 · Score: 1
  
  It's almost as if they were joking... Nah couldn't be.
20. Re:Kids aren't that good at it by instagib · 2011-09-23 23:44 · Score: 1
  
  Awww, poor kids, their dad has no sense of humour.
  Apparently, including the mods. I foresaw that and included a smiley, but it didn't help.
  In fact, having kids seems to correlate with a low sense of humour (or lack of ability to simply ignore a joke they can't laugh about). Sadly, I've seen this in real life as well. I'm pretty sure the GP was able to laugh about the dumbest jokes until approx. 9 years ago.
21. Re:Kids aren't that good at it by tmjva · 2011-09-24 02:13 · Score: 1
  
  Over twenty years ago I was told a story about a ceramic company selling dinnerware such as plates, bowls and dishes and the marketing on the box said "Hand Painted". That the plates were painted by robotic hands obviously wasn't mentioned.
  
  --
  Tracy Johnson
  Old fashioned text games hosted below:
  http://empire.openmpe.com/
  BT
Very nice by Animats · 2011-09-23 04:26 · Score: 2

That's a neat result. I used to work on that problem. Today's solutions use a lot more compute power, but now that's available. Early approaches to this problem worked by treating it as a maze problem in N-dimensional configuration space and running a maze solver. Latoumbe at Stanford was behind a lot of that. That approach became combinatorialy infeasible as N increased. Newer techniques are more like a random greedy search. That works, but the paths aren't all that great. This latest solution seems to improve on random greedy search. That makes sense.
1. Re:Very nice by sgt+scrub · 2011-09-23 06:48 · Score: 1
  
  Is that an N-dimensional matrix search that uses + in the regular expression?
  
  --
  Having to work for a living is the root of all evil.
2. Re:Very nice by Animats · 2011-09-23 08:06 · Score: 1
  
  No, it's an N-dimensional configuration space. This is a variation on the Piano Mover's Problem.
  Visualize getting a point-sized object through a maze. The maze can be treated as a graph, with junctions as graph nodes. All dead end links and closed subgraphs not containing the endpoints can be discarded. What remains contains a usable path. Then you use a path finding graph algorithm. If links have costs, there are ways to find an optimal or (with much less work) a near-optimal solution.
  Now consider getting a round object through a 2D maze. You can just expand all the walls by the radius of the object. This may close some paths. Then proceed as above.
  The next step up is moving a rectangular object through a 2D maze. Turns may be needed to get through tight corners. This is the piano mover's problem. For this, consider a 3D stack of 2D mazes, representing all the possible orientations of the rectangle. Expand the walls as before to make the problem into a graph, then find the path through the graph.
  It's possible to extend this concept to handle arms with joints. The dimensionality of the space increases with the number of joints, but in the end, it's a graph problem. This is where advanced motion planning was in the mid-1980s.
  That approach leads to solutions which work, but are sort of clunky, since velocity and acceleration aren't considered. The graph gets very large as the workspace gets cluttered, and you need full information about the workspace before you can start. It's a reasonable approach for CNC machine tools, but not robots that have to work in a less structured environment.
  So robotics has moved on to techniques which use more randomized search and produce more efficient motion. See the paper from the article.
  Anyway, that's a very brief introduction to motion planning in configuration space.
3. Re:Very nice by SnowZero · 2011-09-23 18:28 · Score: 1
  
  Latoumbe at Stanford was behind a lot of that.
  In a research family tree point of view, he's behind this too. The MIT algorithm is a modification of RRT that instead of using one-connectedness uses k-connectness like PRM. RRT is from LaValle and Kuffner, who were Latoumbe's students at Stanford. PRM is from Lydia Kavraki, who was also Latoumbe's student at Stanford.
  The PR2 video was kind or irksome, because they chose the very first version of RRT (circa 2000) to compare against their new algorithm, rather than a number of variants that became available within a few years (For example ERRT from 2002, or multi-bride RRT from 2006, which are my small contributions to the space). In particular D Ferguson's DRRT from 2006 which would replan iteratively with a cost constraint would have been a natural baseline to use.
  If they had used any of those later variants, in particular any of the ones that implemented path smoothing (considered an absolute necessity since the beginning), you probably could not tell the difference in the results. IOW it was a strawman just like comparing MyNewSort to bubblesort instead of quicksort.
  It is still a contribution though. I saw a very similar algorithm proposed by some Taiwanese researchers (forget the paper) that was also a graph-based extension of RRT, but Karaman and Frazzoli have gone through and done the proofs to verify reasonable properties for their system (ideas are easier than analysis). However rather than the article's statement:
  "By combining two innovative algorithms developed at MIT..."
  it might be more accurate to say:
  "Building on Stanford's RRT and PRM, MIT researchers have combined them in a new algorithm yielding the best properties of both. It can generate efficient paths without excessive memory use or post-processing."
4. Re:Very nice by SnowZero · 2011-09-23 18:29 · Score: 1
  
  oops...
  s/multi-bride RRT/multi-bridge RRT/
I thought that building was too big for Bender... by Anonymous Coward · 2011-09-23 04:27 · Score: 1

http://theinfosphere.org/Robot_Arms_Apartments
Save your mod-down points folks, you know it had to be done.
Sex lives of /.ers improved by mjwx · 2011-09-23 04:31 · Score: 0

Sex lives of many /.ers improved.

Rule 34 once again proven.

--
Calling someone a "hater" only means you can not rationally rebut their argument.
1. Re:Sex lives of /.ers improved by sgt+scrub · 2011-09-23 06:49 · Score: 1
  
  Sorry. I saw the title and thought there was actually data to be found.
  
  --
  Having to work for a living is the root of all evil.
2. Re:Sex lives of /.ers improved by Flyerman · 2011-09-23 08:15 · Score: 1
  
  Netcraft has confirmed this improvement in Slashdotters' sex lives.
bugs by danlock4 · 2011-09-23 05:05 · Score: 1

Darn it, I bought this inexpensive prosthetic arm from a bootlegger/bootarmer somewhere in Asia and sometimes it flails uncontrollably for between 0.5 and 2 seconds. I'm truly sorry it spilled your snack and you missed that scene. May I compensate you for the wasted theater popcorn? FWIW, my wife can't sleep in the same bed with me if I've forgotten to remove it, if that's any consolation.

--
To .sig or not to .sig, that is the question.
Re: sig by camperdave · 2011-09-23 05:13 · Score: 2

The phrase "as hell" has been so over-used that it can mean anything and is therefore meaningless.

The phrase "as hell" has been so over-used that it can mean anything and is therefore meaningless as hell.

--
When our name is on the back of your car, we're behind you all the way!
Bend It Like Bender by Sponge+Bath · 2011-09-23 05:19 · Score: 1

...at the Robot Arms Apartments, you optimal trajectory to high-rise comfort.
it's all about the straw man by cathector · 2011-09-23 05:36 · Score: 1

the algorithm they're so proud of besting seems pretty crappy.
1. Re:it's all about the straw man by Fned · 2011-09-23 05:50 · Score: 1
  
  It only seems that way because you're super-good at it.
  (Apologies if it turns out you're a klutz)
2. Re:it's all about the straw man by cathector · 2011-09-23 06:11 · Score: 1
  
  i guess i don't understand why this robot behaves so bizarrely [with the simpler algorithm] when there are no shortage of videos of robot arms catching tennis-balls and such.
  as an undergrad i wrote a very simple reach-planning algorithm for a robot arm with N joints which seems like it would outperform the simpler of these algorithms.
  i must be missing something.
3. Re:it's all about the straw man by pushing-robot · 2011-09-23 06:42 · Score: 1
  
  Having some understanding of the field, I'm normally pretty forgiving about the problems involved in getting robots to interact in the real world.
  But in this case, I can't get this out of my head.
  
  --
  How can I believe you when you tell me what I don't want to hear?
4. Re:it's all about the straw man by Fned · 2011-09-23 12:17 · Score: 1
  
  That ball catching thing isn't the same thing at all, it tracks the hand and the ball to points in space.
  AFAIK, it can't detect obstacles in its space and figure out the most efficient way to move the hand from point A to point B without any part of the arm hitting any of the obstacles.
  It's the path-finding part that's tricky.
Another MIT Press Release by Anonymous Coward · 2011-09-23 05:49 · Score: 0

Slashdot = MIT PR news feed.
Can't we at least wait for a credible third source to reprint it?
linear algebra^2 by sgt+scrub · 2011-09-23 06:38 · Score: 2

They used linear algebra to teach neural nets to do linear algebra? My brain hurts.

--
Having to work for a living is the root of all evil.
Okay, cool by Anonymous Coward · 2011-09-23 07:28 · Score: 0

Not sure why this was slashdotted. New motion planning algorithms come out every day. This one is just another variant of RRT. To be fair, it is much better than raw RRT, but there are already many much better algorithms than RRT out there -- sbpl, CHOMP, etc.
where are the sources to back up these claims? by Anonymous Coward · 2011-09-23 13:21 · Score: 0

how does this MIT blog-writer know so much about robotics?
My guess is that the statement "most motion-planning algorithms give up on the idea of finding the most efficient path between the robot’s initial state and its goal" has no sources to back it up.
As a matter of fact complex motion algorithms and 'take-out' times were long ago perfected by private companies. The wonderchildren at MIT had nothing to do with it.
"The Feeling of Power" by tmjva · 2011-09-24 02:02 · Score: 1

I suggest this is the start to the long road described in the short story "The Feeling of Power" by Isaac Asimov. In the future someone will re-invent the art of human computation, and manned missiles (and thus human computing of trajectories) will follow.

--
Tracy Johnson
Old fashioned text games hosted below:
http://empire.openmpe.com/
BT
Re: sig by danlock4 · 2011-09-24 04:32 · Score: 1

Excellent revision to my "meaningless-as-hell" .sig!
However, if it's that meaningless, wouldn't it also be as meaningful as hell? I don't want to perpetuate the redundancy of its meaninglessness when it's equally as meaningful. Heh. Oh, ouch. I really need to wake up before I type myself into a lake of logorrhea.
Thanks for the comment :)

--
To .sig or not to .sig, that is the question.
This algorithm has its own cost by Hentes · 2011-09-24 11:22 · Score: 1

This algorithm might be more optimal but is also slower than the alternatives. Electricity is cheap, time is not.
1. Re:This algorithm has its own cost by Anonymous Coward · 2011-09-25 09:48 · Score: 0
  
  You might want to actually read the paper.