PDA

View Full Version : Best biped for RoboCup?



wmccafferty
06-09-2009, 11:24 PM
Hi all,

I'm looking to purchase a (team of) biped with the intention of competing in RoboCup KidSize league (http://www.tzi.de/humanoid/bin/view/Website/WebHome). Obviously, the Nao (http://www.aldebaran-robotics.com/eng/) would be slick, but is a bit expensive, even with the beta program. It's seems like the humanoid bipeds that Trossen offers are rather competitive with each other. Is there one or two that you all feel stand above the rest? I'm looking to augment the choice with stereo vision, such as http://surveyor-corporation.stores.yahoo.net/srblstca.html.

Thoughts, suggestions?

Billy McCafferty
http://devlicio.us

wmccafferty
06-10-2009, 12:22 AM
Ah yes, I almost forget, and I'd like it to have services available for compatibility with Microsoft Robotics Studio.

Thanks for any suggestions you may have,
Billy McCafferty

lnxfergy
06-10-2009, 09:17 AM
The issue here is the payload you need to carry. You want stereo vision, that weighs quite a bit. It would appear you need ~20 minutes of battery life, that weighs quite a bit. As for MSRS support, don't the rules require no "offboard-brains"? If so, do you plan to slap a windows-PC onboard the bot? A windows PC, capable of doing the required vision processes for the game (heck, capable of doing anything really) would probably weigh so much with its required battery power, so as to rule out anything short of an RX-64-based bot. Otherwise, you'd probably be stuck using something like a CMUCam (which it appears a few competitors have greatly extended to make it viable for this competition), at which point you have to ask, why MSRS?

From what I've seen and understand the KidSize league is even more competitive and technologically advanced than the spec league... I doubt any of the platforms in the KidSize league cost much less than the Nao... especially when you consider the costs of developing a platform. Further, I don't think any of the ~$1000 biped kits are going to be plug and play for something like robocup, you still have to add dynamic balance, vision, high-level processing.

-Fergs

wmccafferty
06-10-2009, 10:01 AM
Fergs,

Thanks a lot for the feedback. The hope for MSRS was to have a sophisticated enough simulator, akin to the bipied itself, in order to develop multi-agent software. While I could just go the simulator route altogether, having some hardware to go along with it is a lot of the fun. ;)

Thanks again for the feedback, you make some great points.

Billy

SK.
07-13-2009, 08:30 AM
Hi,

the top RoboCup teams all use custom built robots of 45-60 cm size with Robotis DX-117/RX-28/RX-64 servos. There were/are teams using modified RoboNovas, RoboBuilder and Bioloids, but the weight of onboard processing really hurts these small robots.
Even the Nao is too slow and fragile to compete against the best Humanoid KidSize teams, at least in itīs current state. You might want to contrast videos of the recent final games from Graz:

YouTube - RoboCup 2009 Humanoid League KidSize Semi Final: Dribblers vs Nimbro 1/2

YouTube - Robocup 2009 FINALS - Standard Platform League (part5)


(http://www.youtube.com/user/DarmstadtDribblers#play/user/23B57AD2A2A1D6BD)

xx2747
07-23-2009, 02:54 PM
Have you seen Team Osaka?
They, have, in my opinion, the most stable robots. I don't know anything about the software but...
I think they sell these too. Anyway, here's a video:

YouTube - RoboCup 2008 Humanoid Soccer Final: NimbRo vs. Team Osaka

SN96
07-26-2009, 09:54 AM
Are those Robo One bots autonomous or R/C controlled?
________
Kawasaki ZL1000A (http://www.cyclechaos.com/wiki/Kawasaki_ZL1000A)

lnxfergy
07-26-2009, 10:13 AM
Are those Robo One bots autonomous or R/C controlled?

Robo-One is RC controlled fighting robots, but these videos are of RoboCup -- which are autonomous soccer playing robots.

-Fergs

darkback2
07-26-2009, 10:56 AM
Ok...so this is more of a philosophical question than anything else.

Why the rule against off board brain? With a kid sized robot, the weight of an onboard PC would be a much greater percentage of the total weight than with a larger bot. So for example, a 1.2 pound micro PC would be about 30% of the weight of a kidsize biped, and only 5% of a much larger biped. Of course the numbers are just made up, but you get the point.

So... a smaller bot is at a greater disadvantage over a larger bot in my opinion.

Thoughts?

DB

lnxfergy
07-26-2009, 11:23 AM
DB, it's just the route they want to go. The lower leagues can do offboard computation, but of course the robocup end goal is:

By mid-21st century, a team of fully autonomous humanoid robot soccer players shall win the soccer game, comply with the official rule of the FIFA, against the winner of the most recent World Cup.

Therefore, the end direction they are clearly going is towards bigger and bigger bots -- eventually fullsize humanoids. On-board processing is probably just their idea of "fully autonomous."

-Fergs

SN96
07-26-2009, 01:36 PM
Robo-One is RC controlled fighting robots, but these videos are of RoboCup -- which are autonomous soccer playing robots.

-Fergs

That's impressive if they are autonomous because they do a good job locating the ball and kicking it in the proper direction. They did so well I thought that they had a puppeteer on the sidelines.
________
buy silver surfer vaporizer (http://vaporizers.net/silver-surfer-vaporizer)

lnxfergy
07-26-2009, 02:52 PM
That's impressive if they are autonomous because they do a good job locating the ball and kicking it in the proper direction. They did so well I thought that they had a puppeteer on the sidelines.

Yep, pretty much everyone there is a large University team. I'd imagine that an average annual budget is well north of $150k. Most of those bots are in the $10-20k range a piece, not including R&D work -- and don't forget the fun costs related to travel. You can accomplish quite a bit by throwing bucket loads of money on something.

-Fergs

SK.
07-31-2009, 04:59 PM
Have you seen Team Osaka?
They, have, in my opinion, the most stable robots. I don't know anything about the software but...
I think they sell these too. Anyway, here's a video:

YouTube - RoboCup 2008 Humanoid Soccer Final: NimbRo vs. Team Osaka (http://www.youtube.com/watch?v=iMM_XQXJUUc)
Team Osakaīs robots are/were indeed of very impressive mechanical design, definitely the most streamlined and reliable robots in Humanoid KidSize. They still lost in the finals 07 and 08 against Nimbro though and didnīt take part in RoboCup 2009. Itīs thus hard to say how theyīd fare these days.

SK.
07-31-2009, 05:33 PM
Ok...so this is more of a philosophical question than anything else.

Why the rule against off board brain? With a kid sized robot, the weight of an onboard PC would be a much greater percentage of the total weight than with a larger bot. So for example, a 1.2 pound micro PC would be about 30% of the weight of a kidsize biped, and only 5% of a much larger biped. Of course the numbers are just made up, but you get the point.

So... a smaller bot is at a greater disadvantage over a larger bot in my opinion.

Thoughts?

DB
Well, one of the goals is research in truly autonomous systems, where onboard processing is either desirable or even needed. Examples would be space exploration, many cases of search and rescue robotics or even house robots. Itīs much more convenient to have a self-contained system doing itīs thing, than having to wire your complete house with cameras, radio comms and whatnot. And thatīs not even considering that for many applications latency and (transmission) reliability concerns mean you have to do onboard processing.

So doing onboard processing at KidSize Humanoids was a conscious decision that certainly represents some difficulties, but those are there to be solved (and solving them is the point basically) :wink:.
It might sound a little strange, but despite onboard processing weighting more relative to total weight for KidSize than for TeenSize robots, the former are much more nimble than the latter. Bipedal locomotion is far from "solved" for KidSize robots, but at least thereīs stable and fast locomotion available for them, courtesy of scaling, good available actuators for that scale and physics. TeenSize robots are struggling with walking much more still, due to their size and the difficulties arising from that.
Have a look at the KidSize and TeenSize matches or finals and youīll see a huge difference.

SK.
07-31-2009, 05:45 PM
Yep, pretty much everyone there is a large University team. I'd imagine that an average annual budget is well north of $150k. Most of those bots are in the $10-20k range a piece, not including R&D work -- and don't forget the fun costs related to travel. You can accomplish quite a bit by throwing bucket loads of money on something.

-Fergs
Of course, no RoboCup team could exist without money. But throwing money at robots doesnīt make them autonomous. Thatīs the work of the team members who put a lot of effort, time and sweat into the projects. Most universities canīt afford throwing just money at their projects, so they have to throw motivated team members (mostly students) at them. :happy:

lnxfergy
07-31-2009, 08:56 PM
Of course, no RoboCup team could exist without money. But throwing money at robots doesnīt make them autonomous. Thatīs the work of the team members who put a lot of effort, time and sweat into the projects. Most universities canīt afford throwing just money at their projects, so they have to throw motivated team members (mostly students) at them. :happy:

True, but being in a University setting myself, I find that locating motivated people (students) is typically FAR easier than finding that sort of cash for a competition. The same goes for open source projects out in the real world.

-Fergs

Adrenalynn
08-01-2009, 12:56 AM
Onboard PREprocessing, I'd agree with. Extraterrestrial autonomous rovers have very basic onboard behavior and even more basic run-out programs. The remainder is external processing.

"Wiring a house" ... ? Wow. I'm not sure why anyone would these days, since [at least in California] I haven't been in a house in years that wasn't UNwired. WiFi is incredibly high-bandwidth and low-cost these days. And in that kind of local environment, latency is low enough for even near-realtime behavior.

You just won't be able to carry the processing power on a sub $100,000 biped for many decades, where multi-terraflop desktop computing exists commonly (and reasonably inexpensively) today.

I have difficulty watching any of those biped soccer matches. They're painfully boring for me. It's like watching golfers watching baseball players watching paint dry on growing grass... "Oh! Oh! The ball's sitting there. Here comes the biped! .... here still comes the biped. ... any day now... oh! It's going to kick the ball! It missed. Oh! Here it goes again! Oops, nope, it fell on its shiny metal arse. Oh, it's trying to get back up. There it is! Oh, no, it fell over forwards. Oh! It's back up! It's going to kick the ball! It kicked it! It rolled two inches and the 'bot fell over on top of it."

nagmier
08-01-2009, 02:03 AM
TY! That was the best laugh I've had all week!

SK.
08-01-2009, 02:57 AM
Onboard PREprocessing, I'd agree with. Extraterrestrial autonomous rovers have very basic onboard behavior and even more basic run-out programs. The remainder is external processing.
Correct, right now. It certainly would be advantageous to have more autonomy though, so you don´t have huge latencies or have to wait for a connection.


"Wiring a house" ... ? Wow. I'm not sure why anyone would these days, since [at least in California] I haven't been in a house in years that wasn't UNwired. WiFi is incredibly high-bandwidth and low-cost these days. And in that kind of local environment, latency is low enough for even near-realtime behavior.Yes, but when you do image processing at 30Hz / 640x480 size for example, WiFi quickly doesn´t look incredibly high-bandwidth and low latency (at least if you don´t use compression, which has other disadvantages). And when everyone around you is using it too you can basically forget about transmitting any meaningful amount of larger sensor data.


You just won't be able to carry the processing power on a sub $100,000 biped for many decades, where multi-terraflop desktop computing exists commonly (and reasonably inexpensively) today.Of course, but one will always have to work with what´s available for small autonomous systems. Now if there´d be a 100% reliable, unlimited bandwidth, no latency datalink that works flawlessy through every obstacle that might be a different story.


I have difficulty watching any of those biped soccer matches. They're painfully boring for me. It's like watching golfers watching baseball players watching paint dry on growing grass... As you might have guessed :veryhappy: I´m a bit impartial here. I´d use your description for the RoboCup 2005 KidSize league (and partially the current Standard Platform League), but not really for the 2009 Humanoid KidSize League. Of course your mileage might differ (and it probably does :) ). For me it´s the opposite, I find remote controlled robot soccer rather boring, even if it´s a little faster.

2005 RoboCup Humanoid KidSize (and parts TeenSize):
YouTube - Humanoid League Team NimbRo at RoboCup 2005

2009 RoboCup Humanoid KidSize:
YouTube - RoboCup 2009 Humanoid League KidSize Final: Darmstadt Dribblers vs FUmanoid 1/2


And a great example from 2006 which probably illustrates rather well how viewpoints can differ. For some this footage shows a robot slowly stumbling around the ball (which is true), for others it shows the incredible first autonomous backheel kick by a robot (and it was very surprising he did that ;) ):
YouTube - Humanoid robot 'Bruno' demonstrates heel kick

Adrenalynn
08-01-2009, 03:35 AM
Still all looks like watching paint dry to me...

As you may or may not have read elsewhere - video is my specialty. I've been doing codecs for nearly thirty years. I disagree with your assessment of the state of video. Full D1 uncompressed video is never a requirement for, well, anything. There's not much you'd need to do with video that you couldn't run 16 channels or more over consumer-grade wifi for today, assuming the system is reasonably local and reasonably well designed. My latest work will move 30hz evidentiary quality D1 video from 32 channels over 802.11g with 10-15ms of latency.

Just getting the processing a few feet off the 'bot means mobility and limitless processing power.

>> incredible first autonomous backheel kick by a robot

Even a blind squirrel finds an acorn now and again. ;)

SK.
08-01-2009, 04:41 AM
Still all looks like watching paint dry to me...
Well you have some blazing fast drying paint over there ( and being in sunny California helps I suppose ) :veryhappy:


As you may or may not have read elsewhere - video is my specialty. I've been doing codecs for nearly thirty years. I disagree with your assessment of the state of video. Full D1 uncompressed video is never a requirement for, well, anything. There's not much you'd need to do with video that you couldn't run 16 channels or more over consumer-grade wifi for today, assuming the system is reasonably local and reasonably well designed. My latest work will move 30hz evidentiary quality D1 video from 32 channels over 802.11g with 10-15ms of latency.Have read only about your LIDAR efforts, very impressive IMHO. I got some RoboCup WiFi experience, itīs not much fun if you sometimes have 200+ APs in one hall.
Forgive my ignorance, but how does 16 channel WiFi work? Iīm a bit baffled because default 802.11g has the well known 13 overlapping channels and now Iīm wondering if those are related to your 16 channels or if they are something completely different (sure sounds like it).


Just getting the processing a few feet off the 'bot means mobility and limitless processing power.Of course, it depends on what youīre trying to achieve. If you want faster robots ASAP thatīs surely the way to go. If you want research in small autonomous systems that can act autonomously without infrastructure, you prefer onboard processing.
By the way, the RoboCup Small Size League sounds more like it suits your interests. They use offboard processing with a 60Hz Cam above the playing field and only send motion commands to the robots. The ball oftentimes moves so fast that one can hardly track it. They started an experimental humanoid division using Kondos with Markers as their heads.

SSL footage 2009
YouTube - Skuba robocup 2009


SSL Humanoid:
YouTube - SSL-H in RoboCup Japan Open 2009


Even a blind squirrel finds an acorn now and again. ;)Well those blind squirrels find them faster and faster each year, so thereīs hope :)

Adrenalynn
08-01-2009, 06:16 AM
The 16 channels are channelized/packetized over a single wifi channel. Learning to cooperate would probably be helpful in that hall. ;) There's at least a few million independently addressable "channels" there before we even get into a discussion about IPv6. Backbone fabric on commercial switches can run up 30+ gigabits before we need to stop shopping on eBay. Excellent quality full D1 video can be sent around in the neighborhood of 300kbit/sec these days. ATSC/HD around 3-3.5 times that. Putting 64, even 128, possibly 256 cameras on a single WiFi network isn't out of the question. I have commercial installations with 45 WiFi cameras.

You know, I'm sure this will get me strung-up around these parts, but... If bipedal motion is so desirable, why do we build wheeled and tracked vehicles to transport ourselves and perform our work? If artificial bipeds were all that hot for performing tasks, they'd have progressed a lot more in the last few thousand years.

Bipeds are inherently slow, unstable, lack balance, lack mobility, ... From my perspective, research into bipedal motion is mental masturbation. Fun, but not really going to advance the universe.

Of course, someone will parade out the stairs argument, and then I'll have to parade out the videos of tracked and wheeled vehicles blasting up stairs at ungodly speeds. Then someone will parade out the uneven terrain and I'll have to parade out the rock crawlers climbing over boulders larger than refrigerators, ... I should probably just let the topic go - it's been done.

Thanks for the compliment re: lidar! Failed experiment though.

SK.
08-01-2009, 08:36 AM
Interesting regarding the Video WiFi stuff, you learn something new every day :)


You know, I'm sure this will get me strung-up around these parts, but... If bipedal motion is so desirable, why do we build wheeled and tracked vehicles to transport ourselves and perform our work? If artificial bipeds were all that hot for performing tasks, they'd have progressed a lot more in the last few thousand years.Go back some 110 years and you could say the same about airplanes. Technology isnīt there yet, thatīs IMHO the reason why we donīt see bipeds performing some tasks. Iīm no dogmatic biped proponent, but the advantages of having a roughly human size, roughly human shaped robot using human-like locomotion (and maybe communication) for performing tasks in environments built for humans are pretty obvious.
Of course you can also send tracked vehicles around stairs, but those likely wonīt look too good after a tracked vehicle went over them the 500th time. So my guess is that at least for operating in environments built for humans, bipeds are the natural choice, once they work well enough (whenever that may be). That being said, outside of this scenario, tracked and wheeled vehicles are probably more often than not the better choice than bipeds.


Bipeds are inherently slow, unstable, lack balance, lack mobilityOf course, todayīs bipeds are far from practical. Itīs insane how much energy gets wasted due to rigid actuators and links, springless designs etc.
I expect that instability will get solved with better actuators and sensors in the future. Modern Quadrotor UAVs (or fighter jets, for that matter) are also inherently instable designs, but they work pretty well, courtesy of digital filtering and ever better and smaller MEMS sensors.
Once you have (close to) human-like mobility I donīt think one can complain about lack of mobility. You then have more or less a "Jack of all Trades, Master of None" robot. Just like what we humans are, locomotion wise. For many tasks ( as evidenced when one of us uses a bicycle for speeding along a road ) biped locomotion certainly isnīt the most efficient way of moving around.

darkback2
08-01-2009, 09:08 AM
I've said this one before...I think we get stuck on paths...especially when we start down them before we can see the end. 40 years ago the idea of a humanoid robot, or a single robot that could do all sorts of stuff for you (as in replace the maid) was appealing because computers/robot brains were so expensive. You would need one robot that can do a large number of tasks...so...it would be best if it looked like us, and could do all of the things we do.

Brains are cheap now. Both in terms of weight and money. You can put brains in anything. more than one brain...and so on. For example, why would I need a robot to bring me coffee, if I can have an automatic coffee maker in every room of my house? Of course you could argue...who would deliver the beans...I suppose a rover would be better at that...or a set of tubes through my house...

Another example would be that the same robot that brought me coffee could also drive me to work. But why have a robot take up a seat that could be used by a human when there is no need for a robot to be in the car at all...why not just make the car the robot. Again...brains are cheap.

As for robocup...I think its more than just masturbation. There are other applications to the technological advances that are needed for a team of robots playing soccer, and the challange of doing it with bipeds just adds to the challenge. Who knows, maybe prosthetic legs that respond to thought control will be available as a result, or swarms of robots that can cooperatively manage a forest. Who knows until we try.

DB

DB

SK.
09-23-2009, 03:38 PM
Back on topic, the slides of a presentation recently held at the RoboCup Mexican Open ( http://www.torneomexicanoderobotica.org.mx/tmr/ ) about this yearīs best humanoid are available online:
http://www.prisma-forum.info/presentaciones/Darmstadt_Dribblers_Presentation_2009-09-11.pdf

Topic is "Science, Technology, and Team of the 2009 RoboCup Champion" and probably worth a look if one is interested in the RoboCup Humanoid League.