i based my post on my experience with two companies that actually stopped offering products for samples after amateur robotics clubs started sampling their products en masse. it does happen. (though my post was an overreaction, given. funny what you can write at 4am...)
however these were micros and connectors (some of which are pretty expensive per-unit), not fets/amps and such, so thanks for your POV and so on:)
The fact that the writeup included the bit about samples is really kinda dumb. Moderation is key. Samples are great but if every idiot starts sampling everything (which i'm sure will be a side effect of the present article) companies will stop sampling or make it more difficult.
in general, the state of slashdot is shameful these days. i dont have a solution (aside from simple obvious things like submission moderation, etc)... maybe i've just changed enough that it isnt the place for me anymore. which is a shame. cause from my POV slashdot aspires to be about Cool Things. the latest microsoft bug isnt a cool thing. it isnt news. (to adapt what John Stewart said about a transmission from Hussein).
and all of this Geek Nerd etc shit. I think the US population is nuts about trying to group people (including themselves!) into groups of like scales. I havent seen anything like it anywhere else (i live in US and have lived in other places).
anyway what was i gonna say? oh yeah: to anyone who reads it - if you sample, please, PLEASE sample in moderation so that people that actually build prototypes and such (like *this) continue to have this wonderful resourse availible.
These guys at stanford have done some really amazing stuff that's directly related. Except that they has literally dozens of cameras (as seen in their ppt), and their research seems to concentrate on multifocal image reconstruction (see ppt slides, presentation is quite good)
The SDR4X aka QRIO is also expected to sell for ~80K, if it is ever becomes commercially availible. This is an estimate i got from someone close to the project 2 years ago.
Re:sensors and subprocessors.
on
Animal Robots
·
· Score: 1
2 accelerometers can be used to detect and measure the rate of rotation (error is pretty bad at this point, but its possible)
gyro's do a very good job at measuring rotation. they are not that expensive either
you can get free (sample) and rather good mems accelerometers from A/D. gyros are about $60 for the ones i've used.
also can use magnetometers or full IMU for very nice closed-box intertial sensing. those cost quite a bit.
packbot was at the 2004 RoboCup american open. its pretty cool. can climb stairs and the cool arm-drive system seems to handle litter on the floor and obstacles really well
its easy to recognize a table from a chair. that statement is meaningless. now when you dont know that its one of the two - thats different
this is someone from cornell robocup speaking. we leave for portugal (international competition) tomorrow. wish us luck.
i cant find the cheap linx modules i've seen before, but there are some really basic laipac ones for $10 at www.sparkfun.com. (granted, you have to do packetizing/filtering yourself)
This is also the place where you can get really cheap atmel avr programmers and a $4 surfboard for atmels atmega128 (a tqfp-44 chip!)
forgot: use either sonar or IR rangers for either determining where you are or simply avoiding obstacles. you can also use feather switches as a last resort
a) get two|three|four big DC motors (gearing probably a good idea but you could probably find two large ones at a scrap yard somewhere) b) get a breadboard and a micro (i recommend atmegas - chip is ~$5 + $9 for programmer + compiler (WinAVR) is free/opensource c) get a 5v linear regulator, some optoisolators, and (easiest) get some cheap high-amp MOSFET's (remember that you can always put them in parallel for pretty much unlimited current) d) get a pair of crappy linx RF serial modules (cheap, easy to set up) e) use 9v battery for digital power supply. use the biggest possible lead acid for motor power supply f) go nuts!
(if you dont know how to hook these up together, you should first experiment with simpler circuits. epanorama is a good source of info)
oh boy, to inject some super-secret pseudo-knowledge here...
optical systems work ok in desert-like environments. they work somewhat well for picking out obvious obstacles (bushes, rocks, sharp cliffs), but definitely can not guide a vehicle alone.
stereo vision is just the beginning. a whole lot more here than "two web cams". that does NOT mean that you won't be successful, just start researching... now.
some machine-vision keywords to look for: motion sensing, vector-based frame comparisons (look at how frames change and try to determine what the hell is going on from that... takes 1 camera for very rough 3d sensing)
LADAR and such help here alot, as they give you full 3d information (the end thing that you really want), but they have their own problems: expensive and SLOW. you're not gonna drive at 50 mph with those. not even much slower
so, you have to basically get all the damn sensors you can get and try to get as complete of a picture as possible
btw, the coolest things i've seen that are relevant (sensors) are cameras that produce depth pictures instead of color pictures. maybe this will help someone
(perhaps i will be in this in an "unnamed team", maybe not..) (remember... have fun.. the competitive side is BS, all DARPA wants is a more PR-friendly way to kill) - v14 / iegres
side note: in many universities MSVS, along with pretty much every major MS product except office are availible for free (legally) (COMPLETELY free dl) through MSDNAA
bad form to reply to self, etc etc, but just wanted to put the final nail into this coffin (err....):
"The gyros are polled at 100Hz, which is overkill considering the height of the robot's CG. With six gyros churning at 100hz, a lot of mission-critical bandwidth is required, so I placed the gyros on their own token-ring controller that is accessible only to the balance and watchdog CPUs. "
100hz is SLOW for low-level feedback control. we run 4-wheel omni-dir robots and they go at 300 hz... i would think that you need an even higher control frequency (and lots of sensors/fast actuators!) when you're balancing a humanoid.
the robots center of gravity has nothing to do with this. also you would probably have more than 6 gyros for this many limbs... and you'd also have accelerometers, etc. a token ring has NOTHING to do we gyro/accel-type sensors. in fact all a gyro/acc requires for output is usually a single wire connection (analog or duty cycle output)
also, humanoids are REALLY difficult to keep balanced, not even speaking of through doing some actions (stopping cars?????). oh yeah, and motors with torque high enough to actually support that thing either a) don't exist or b) are too slow or c) REALLY expensive. take a look at the sony SDR-4X robot ( i think it got renamed recently) and its joints. compare to Honda's ASIMO thats a few years older. etc etc...
Also listen to the interview- especially the "leg is a pendulum, so we used a fly-wheel" crap. I have some (limited) experience with humanoid-type robotics, but this guy straight BS'es through both interviews. its pretty amusing through.
AFAIK, they use a custom versioning/simultaneous build/test tool. its not really reasonable to expect them to use source safe for the scope of their main projects...
btw, i used to use source safe (for a bit). i didn't really use it that much, but i can't complain. certainly it was no worse than CVS, which is still lacking critical features ("prune empty directories" anyone?).
i'm really glad subversion is going gold (in the way software like this goes gold...). i look forward to using it.
I also work on a university research project (robocup.mae.cornell.edu). We had no problem getting an educational license from Qt. Just apply for an edu license and they'll give you links to the appropriate binaries. (though it may take a bit of time - week to 2 weeks)
the only difference from the GPL/commercial versions is you don't get the full source. otherwise, exactly all the features of the latest version.
some publishers/organizations already let you do it, providing you're a subscriber. example: ACM or IEEE (together thats a database of >200 journals over 15 years!)
of course, you do need to be a subscriber...
i agree that a universal abstract/full text search would be nice, but many things have to happen for that to exist. possibly some kind of (UNIVERSAL) database where authors submit their published papers would be good. freeware, preferably, too. and with the customary free-publishing lag, it shouldn't interfere with the existing publications too much.
just TRY to sing that:-). the "llllllll" and the "eeeee" might work (though deleteeeeee might sound a bit weird...), but how are you going to sing "altttttttttttttttt"?;-)
we, believe it or not, tried bluetooth for our vision->robot communication. my advice: try something simpler to just get your basic system running.
for example, we used Radiometrix RPC's for 4 straight years.. they are really limited but everyone uses them because their problems are well known in the league. all i can say is, good luck.. and drop us (site) an email if you have specific problems..
yo yo. sergei of the current team (cs) here. we're very sorry for the embarrasing game. we were halfway done porting the 2002 ai into our brand new framework and a) our defense was not tested b) we had not offense c) our vision was messed up (old 2002 vision...)
we promise a much better showing at italy:) -sergei
Fira sounds very much like exactly like RoboCup. now, i don't know which one came first or if they're in competition or anything, but i fail to see why you thing it has more engineering merit. FYI, RoboCup also has quite a few leagues: AIBO, SSL, Rescue, Simulation (at CMU they had a REALLY cool interface), and a bunch more). so, if you can provide any specific examples of rules, it would be much appreciated.
The second paragraph is a necessity for any robot soccer-system of this type. For wireless, some SSL teams use 802.11b (all AIBO's use it AFAIK). Many SSL teams use wireless modules on the 433/418 frequencies. H-bridges or such (i.e. pairs of MOSFETs) are unavoidable. vision is too (working on that now:) ).
i completely agree with you on the last statement in p2. these are great multi-disciplinary projects.
this would make the game completely impossible for new teams. at the US Open only CMU and Cornell even had working systems in the SSL F-180 league...
RoboCup is still "just getting started". it takes alot of resourses/time/money to build a working system, and we want to ENCOURAGE new teams to enter, not discourage them.
there are many other rules that would be implemented if this weren't the case. Full autonomy is a biggie. there are other ones such as rules on touching and so fourth. -sergei/Cornell BigRed'03
sorry for losing against cmu, we were having vision trouble:(.. on the upside i must say the bots were in a pretty good shape mechanically and electrically. there were 2 bugs in our defense and every one except one (maybe 2) goals scored against us was because of those bugs...
its quite annoying to read slashdot and see everying misquoted and plain false facts
anyway, the AIBO's certainly win by sheer cuteness. unfortunately, those other F180 teams didnt really do too well:(.
in the end though the experience of having our completely new team go over there and set everything up and just talking to the other teams far far outweights any competitiveness. to be honest i really wish the other teams did better... it was a shame that small problems bogged so many systems down -sergei
yes yes, its Micro$oft, but i thought this was a really cool and insightful lecture. The guy shows off keyboards with additional and very much radical components, a gyro/accelerometer-enchanced PDA, among other things:
Slashdot Mirror
i based my post on my experience with two companies that actually stopped offering products for samples after amateur robotics clubs started sampling their products en masse. it does happen. (though my post was an overreaction, given. funny what you can write at 4am...)
:)
however these were micros and connectors (some of which are pretty expensive per-unit), not fets/amps and such, so thanks for your POV and so on
The fact that the writeup included the bit about samples is really kinda dumb. Moderation is key. Samples are great but if every idiot starts sampling everything (which i'm sure will be a side effect of the present article) companies will stop sampling or make it more difficult.
in general, the state of slashdot is shameful these days. i dont have a solution (aside from simple obvious things like submission moderation, etc)... maybe i've just changed enough that it isnt the place for me anymore. which is a shame. cause from my POV slashdot aspires to be about Cool Things. the latest microsoft bug isnt a cool thing. it isnt news. (to adapt what John Stewart said about a transmission from Hussein).
and all of this Geek Nerd etc shit. I think the US population is nuts about trying to group people (including themselves!) into groups of like scales. I havent seen anything like it anywhere else (i live in US and have lived in other places).
anyway what was i gonna say? oh yeah:
to anyone who reads it - if you sample, please, PLEASE sample in moderation so that people that actually build prototypes and such (like *this) continue to have this wonderful resourse availible.
These guys at stanford have done some really amazing stuff that's directly related. Except that they has literally dozens of cameras (as seen in their ppt), and their research seems to concentrate on multifocal image reconstruction (see ppt slides, presentation is quite good)
Link (has cool results links)
The SDR4X aka QRIO is also expected to sell for ~80K, if it is ever becomes commercially availible. This is an estimate i got from someone close to the project 2 years ago.
2 accelerometers can be used to detect and measure the rate of rotation (error is pretty bad at this point, but its possible)
gyro's do a very good job at measuring rotation. they are not that expensive either
you can get free (sample) and rather good mems accelerometers from A/D. gyros are about $60 for the ones i've used.
also can use magnetometers or full IMU for very nice closed-box intertial sensing. those cost quite a bit.
packbot was at the 2004 RoboCup american open. its pretty cool. can climb stairs and the cool arm-drive system seems to handle litter on the floor and obstacles really well
its easy to recognize a table from a chair. that statement is meaningless. now when you dont know that its one of the two - thats different
this is someone from cornell robocup speaking. we leave for portugal (international competition) tomorrow. wish us luck.
bah, no one will read this. go build something.
i cant find the cheap linx modules i've seen before, but there are some really basic laipac ones for $10 at www.sparkfun.com. (granted, you have to do packetizing/filtering yourself)
This is also the place where you can get really cheap atmel avr programmers and a $4 surfboard for atmels atmega128 (a tqfp-44 chip!)
forgot:
use either sonar or IR rangers for either determining where you are or simply avoiding obstacles. you can also use feather switches as a last resort
a) get two|three|four big DC motors (gearing probably a good idea but you could probably find two large ones at a scrap yard somewhere)
b) get a breadboard and a micro (i recommend atmegas - chip is ~$5 + $9 for programmer + compiler (WinAVR) is free/opensource
c) get a 5v linear regulator, some optoisolators, and (easiest) get some cheap high-amp MOSFET's (remember that you can always put them in parallel for pretty much unlimited current)
d) get a pair of crappy linx RF serial modules (cheap, easy to set up)
e) use 9v battery for digital power supply. use the biggest possible lead acid for motor power supply
f) go nuts!
(if you dont know how to hook these up together, you should first experiment with simpler circuits. epanorama is a good source of info)
estimated cost: $60 + existing lawn mower
oh boy, to inject some super-secret pseudo-knowledge here...
optical systems work ok in desert-like environments. they work somewhat well for picking out obvious obstacles (bushes, rocks, sharp cliffs), but definitely can not guide a vehicle alone.
stereo vision is just the beginning. a whole lot more here than "two web cams". that does NOT mean that you won't be successful, just start researching... now.
some machine-vision keywords to look for: motion sensing, vector-based frame comparisons (look at how frames change and try to determine what the hell is going on from that... takes 1 camera for very rough 3d sensing)
LADAR and such help here alot, as they give you full 3d information (the end thing that you really want), but they have their own problems: expensive and SLOW. you're not gonna drive at 50 mph with those. not even much slower
so, you have to basically get all the damn sensors you can get and try to get as complete of a picture as possible
btw, the coolest things i've seen that are relevant (sensors) are cameras that produce depth pictures instead of color pictures. maybe this will help someone
(perhaps i will be in this in an "unnamed team", maybe not..)
(remember... have fun.. the competitive side is BS, all DARPA wants is a more PR-friendly way to kill)
- v14 / iegres
side note: in many universities MSVS, along with pretty much every major MS product except office are availible for free (legally) (COMPLETELY free dl) through MSDNAA
bad form to reply to self, etc etc, but just wanted to put the final nail into this coffin (err....):
"The gyros are polled at 100Hz, which is overkill considering the height of the robot's CG. With six gyros churning at 100hz, a lot of mission-critical bandwidth is required, so I placed the gyros on their own token-ring controller that is accessible only to the balance and watchdog CPUs. "
100hz is SLOW for low-level feedback control. we run 4-wheel omni-dir robots and they go at 300 hz... i would think that you need an even higher control frequency (and lots of sensors/fast actuators!) when you're balancing a humanoid.
the robots center of gravity has nothing to do with this. also you would probably have more than 6 gyros for this many limbs... and you'd also have accelerometers, etc. a token ring has NOTHING to do we gyro/accel-type sensors. in fact all a gyro/acc requires for output is usually a single wire connection (analog or duty cycle output)
also, humanoids are REALLY difficult to keep balanced, not even speaking of through doing some actions (stopping cars?????). oh yeah, and motors with torque high enough to actually support that thing either a) don't exist or b) are too slow or c) REALLY expensive. take a look at the sony SDR-4X robot ( i think it got renamed recently) and its joints. compare to Honda's ASIMO thats a few years older. etc etc...
Also listen to the interview- especially the "leg is a pendulum, so we used a fly-wheel" crap. I have some (limited) experience with humanoid-type robotics, but this guy straight BS'es through both interviews. its pretty amusing through.
different OS'es for different tasks:
winCE.net (4.2):
smallest possible: ~500 kb (yes, KILOBYTES)
largest possible: ~32 megs
oh, and the 32meg build is probably everything quite a few people ever use... including IE, minesweeper, office (well, word...), etc
in other news, this article is completely useless. thanks.
-sergei
AFAIK, they use a custom versioning/simultaneous build/test tool. its not really reasonable to expect them to use source safe for the scope of their main projects...
btw, i used to use source safe (for a bit). i didn't really use it that much, but i can't complain. certainly it was no worse than CVS, which is still lacking critical features ("prune empty directories" anyone?).
i'm really glad subversion is going gold (in the way software like this goes gold...). i look forward to using it.
I also work on a university research project (robocup.mae.cornell.edu). We had no problem getting an educational license from Qt. Just apply for an edu license and they'll give you links to the appropriate binaries. (though it may take a bit of time - week to 2 weeks)
the only difference from the GPL/commercial versions is you don't get the full source. otherwise, exactly all the features of the latest version.
good luck
-sergei
some publishers/organizations already let you do it, providing you're a subscriber. example: ACM or IEEE (together thats a database of >200 journals over 15 years!)
of course, you do need to be a subscriber...
i agree that a universal abstract/full text search would be nice, but many things have to happen for that to exist. possibly some kind of (UNIVERSAL) database where authors submit their published papers would be good. freeware, preferably, too. and with the customary free-publishing lag, it shouldn't interfere with the existing publications too much.
|CONTROLLLLLLLLLLLLLLLLLL!
:-). the "llllllll" and the "eeeee" might work (though deleteeeeee might sound a bit weird...), but how are you going to sing "altttttttttttttttt"? ;-)
| ALTTTTTTTTTTTTTTTTTTTTTT!
| DELETEEEEEEEEEEEEEEEEEEE!
just TRY to sing that
we, believe it or not, tried bluetooth for our vision->robot communication. my advice: try something simpler to just get your basic system running.
.. and drop us (site) an email if you have specific problems..
/me watches this comment be ignored.)
for example, we used Radiometrix RPC's for 4 straight years.. they are really limited but everyone uses them because their problems are well known in the league. all i can say is, good luck
-sergei (cornell robocup 2003,2004)
(sigh...
yo yo. sergei of the current team (cs) here. we're very sorry for the embarrasing game. we were halfway done porting the 2002 ai into our brand new framework and
:)
a) our defense was not tested
b) we had not offense
c) our vision was messed up (old 2002 vision...)
we promise a much better showing at italy
-sergei
Fira sounds very much like exactly like RoboCup. now, i don't know which one came first or if they're in competition or anything, but i fail to see why you thing it has more engineering merit. FYI, RoboCup also has quite a few leagues: AIBO, SSL, Rescue, Simulation (at CMU they had a REALLY cool interface), and a bunch more). so, if you can provide any specific examples of rules, it would be much appreciated.
:) ).
The second paragraph is a necessity for any robot soccer-system of this type. For wireless, some SSL teams use 802.11b (all AIBO's use it AFAIK). Many SSL teams use wireless modules on the 433/418 frequencies. H-bridges or such (i.e. pairs of MOSFETs) are unavoidable. vision is too (working on that now
i completely agree with you on the last statement in p2. these are great multi-disciplinary projects.
-sergei / Cornell BigRed'03
this would make the game completely impossible for new teams. at the US Open only CMU and Cornell even had working systems in the SSL F-180 league...
RoboCup is still "just getting started". it takes alot of resourses/time/money to build a working system, and we want to ENCOURAGE new teams to enter, not discourage them.
there are many other rules that would be implemented if this weren't the case. Full autonomy is a biggie. there are other ones such as rules on touching and so fourth.
-sergei/Cornell BigRed'03
hey len :)
:(.. on the upside i must say the bots were in a pretty good shape mechanically and electrically. there were 2 bugs in our defense and every one except one (maybe 2) goals scored against us was because of those bugs...
:(.
sorry for losing against cmu, we were having vision trouble
its quite annoying to read slashdot and see everying misquoted and plain false facts
anyway, the AIBO's certainly win by sheer cuteness. unfortunately, those other F180 teams didnt really do too well
in the end though the experience of having our completely new team go over there and set everything up and just talking to the other teams far far outweights any competitiveness. to be honest i really wish the other teams did better... it was a shame that small problems bogged so many systems down
-sergei
same reason US children are "forced" NOT to recite the pledge of allegiance.
(this comes from frustration with about a million such headlines)
yes yes, its Micro$oft, but i thought this was a really cool and insightful lecture. The guy shows off keyboards with additional and very much radical components, a gyro/accelerometer-enchanced PDA, among other things:
4
http://murl.microsoft.com/LectureDetails.asp?90