#robotics Logs

Dec 20 2011

06:55 norbi i have to create a project in robotics for my school contest
06:55 norbi i need to come with an idea
06:56 norbi until 21 december, to sign up with
06:56 norbi can you guys help me? i dont have much ideas
07:15 charolastra well, what do you like?
07:16 charolastra norbi: ^
07:57 norbi charolastra: sry i was in the kitchen, i had guests
07:57 norbi really sorry, now im back
07:58 charolastra no problem
07:58 norbi charolastra: anything thats scientific, complicated and interesting
07:59 norbi charolastra: to be honest i really want to win this contest, i had win php programming contests
07:59 norbi and others
07:59 norbi but in robotics didnt had the chance to win yet
07:59 charolastra no, that's to broad; name one thing that you like doing or that's usefull
07:59 norbi i would really like to do something nice
08:00 norbi charolastra: i like programming, avr mcu, humanoid robots, ai
08:00 norbi this is what i like
08:00 charolastra yes, but you won't win anything this way
08:01 norbi charolastra: i dont understand what you mean, maybe my english is the problem:P
08:02 charolastra excuse my rude example but i'm thinking along this lines: you like masturbation? then build a robot that does that for you; you will get ritch and win contests; but of course you have to choose another topic ..... like ... a pill dinspenser for old people
08:03 charolastra they need to take specific medications at a specific time, why not let a robot handle that?
08:04 norbi charolastra: ah you mean, some tematic
08:04 norbi some task for the robot
08:05 norbi the problem is i dont know yet in what would be the robots main task
08:07 norbi and my components doesn't fit to some big robot, i have low amperage motors, some mcu boards, accmeter, adc module, anything thats TTL, but nothing that involves highest voltages
08:07 charolastra yes, that's why i think the other way around; find a trivial task that noone thought about automating and then build the robot
08:08 norbi charolastra: that is a good aproach
08:08 norbi any ideas though?:P
08:08 norbi for something more little?
08:11 charolastra it's a regular school?
08:11 charolastra do you still write on blackboards?
08:11 norbi charolastra: no
08:11 norbi its not regular
08:11 norbi its university
08:11 norbi :P
08:12 charolastra ah, so little blackboard use and no need for a blackboard cleaning robot :(
08:12 norbi :D
08:17 charolastra is it supposed to be mobile?
08:18 norbi not necesarilly, but yea it can be
08:19 norbi i have wheeels, motors at home
08:19 norbi but small ones
08:19 norbi at size of an rc car
08:19 norbi but high torgue, so slow moving
08:31 charolastra well, can't come up with anything original ...
08:33 norbi charolastra: not necesarilly original
08:33 norbi if there is some good idea, but it was done in past, then maybe it would be good
08:33 norbi i dont think i can reinvent the weel yet
08:34 charolastra i don't know ... a toilet paper dispensing robot ... lol
08:34 norbi loool
08:34 norbi :)))
08:36 charolastra an internet conrtrolled robot that can play fetch with you dog from work ...
08:40 norbi charolastra: this is getting closer
08:40 norbi robot controlled via internet is not a problem
08:40 norbi but the playing with my dog wouldnt be a good idea:P my dog would kill him in a sec
08:41 charolastra it might be if latency is an issue
08:41 charolastra hehe
08:41 charolastra i was thinking about urban dogs
08:41 charolastra small ones
08:42 norbi hehe
08:42 norbi i have 5 siberian huskies
08:42 charolastra haha
08:42 norbi i think they would kill it in a sec, and wanting more:D
09:35 thylanefr okay #robotics
09:35 thylanefr the LIDAR on google driverless cars is $60 thousand.
09:35 thylanefr We can do it cheaper.
09:35 thylanefr That is our mission as roboticists.
09:36 thylanefr I think we can do the same thing far cheaper and we can do it with binocular disparity algorithms and lots of fast cameras.
09:36 thylanefr There is a non-linear scaling rule of the number of cameras.
10:21 thylanefr http://www.youtube.com/watch?v=9Y4RQVpp-BY
10:24 thylanefr http://www.youtube.com/watch?v=XejNctt2Fcs&feature=related
10:27 w0lfsp1d3r|2 rhylanefr: on a much smaller scale, do you think it is possible to do a spider-like stereo video using the CMUcam?
10:28 thylanefr i don't know
10:28 thylanefr what does spider-like mean
10:29 w0lfsp1d3r|2 spider robot; someone made one you can control with a iPhone or Android OS device
10:29 w0lfsp1d3r|2 like a jumping spider (without the jumping)
10:30 w0lfsp1d3r|2 it just walks
10:30 w0lfsp1d3r|2 it moves pretty life-like though
10:31 w0lfsp1d3r|2 i forget the name; i have a video from YouTube on my usb hard drive :-\
10:31 w0lfsp1d3r|2 HR Hexapod or something
10:32 thylanefr so you want to utilize the object-rec of two CMUcams?
10:32 w0lfsp1d3r|2 yes
10:34 w0lfsp1d3r|2 the main goal is to make a laser tag robot using real diode lasers
10:34 w0lfsp1d3r|2 i'm doing a simple wheeled version first
10:36 w0lfsp1d3r|2 if need be, i could use my netbook as a backend over wifi or bluetooth transceiver
10:36 thylanefr i think i would need a clearer idea of the use of various algorithms at distances versus very close objects.
10:37 thylanefr At large distances, it might be better to use optical flow
10:37 thylanefr nearby objects are better for binocular disparity
10:39 w0lfsp1d3r|2 the laser tag would be run in an old school gymnasium
10:39 w0lfsp1d3r|2 larger than a tabletop; smaller than outdoors
10:40 w0lfsp1d3r|2 the onboard processor would be an Arduino
10:40 w0lfsp1d3r|2 which can network to something else if need be...
10:41 thylanefr w0lfsp1d3r|2 So im thinking that depending on the size of the robot's body itself, you are limited by how wide the cameras will be from each other, which in turn limits the range of the binocular disparity
10:41 w0lfsp1d3r|2 i think it would be cool to make a mini-grid computer out of Arduinos :-)
10:42 w0lfsp1d3r|2 the robot would be about the size of an RC car
10:42 thylanefr i can't imagine binocular vision being useful beyond 2 meters with that distance
10:45 w0lfsp1d3r|2 to simpify the robot can you just move the platform instead of moving the cameras on the platform?
10:45 w0lfsp1d3r|2 spiders don't seem to move their eyes...
10:45 thylanefr real spiders?
10:46 w0lfsp1d3r|2 yes
10:46 thylanefr they are using optical flow of some sort
10:46 w0lfsp1d3r|2 i'll have to put that into Wikipedia...
10:47 thylanefr plus their brains actually recognize objects. So they have an 'idea' of how large something is, giving them automatic depth cues
10:47 w0lfsp1d3r|2 ok that is helpful
10:48 w0lfsp1d3r|2 for the control computer i wanted to use a Neural Network based Subsumption Architecture
10:49 w0lfsp1d3r|2 the final hardware would be written in SystemC
10:50 w0lfsp1d3r|2 for now just plain C
10:50 w0lfsp1d3r|2 or Processing for the Arduino
10:51 w0lfsp1d3r|2 what is Optical Flow?
10:51 thylanefr subsumption is useful when you are not using fancy deep-processing methods like localization from vision.
10:52 thylanefr Subsumption was developed for robots with several simple sensors in conjunction.
10:52 thylanefr It sounds like you WILL be using fancy localization schemes
10:53 w0lfsp1d3r|2 hmmm...output from the camera goes into the NN
10:53 thylanefr Well you don't want to do that
10:54 w0lfsp1d3r|2 the camera has its own preprocessing
10:54 w0lfsp1d3r|2 it outputs in hex
10:54 w0lfsp1d3r|2 a bitmap
10:55 thylanefr I can't see how subsumption is going to help here if you have a camera attached.
10:56 thylanefr optical flow is just an algorithm where if you are moving quickly through space nearby objects move faster than distant ones
10:56 w0lfsp1d3r|2 oh ok
10:57 w0lfsp1d3r|2 lunchtime; thank you for the talk :-)
10:57 w0lfsp1d3r|2 it was helpful
10:58 w0lfsp1d3r|2 will you be here for a while?
10:58 thylanefr nope sorry im leaving :)
10:58 w0lfsp1d3r|2 oh ok
13:17 thylanefr We need to start hammering down, in a real way, the plusses and minuses of replacing LIDAR with multiple cameras+binocular disparity
13:17 thylanefr We do know that human beings drives cars every day with two little eyes on their face.
13:20 Steffanx And an advanced brain thylanefr ..
13:21 Steffanx *complex
13:56 norbi for an rc car to use binocular disparity algorithm, that needs 2 cmos cameras or just one?
13:57 norbi im thinking to do that for my school project
14:02 amee2k program it to run a slalom around a row of traffic cones
14:02 amee2k extra style points for drifting
14:06 norbi hehe
14:06 norbi does it need 2 or just one camera?
14:06 amee2k i'd point at two because you can't judge distance with just one
14:06 amee2k well, not algorithmically anyway
14:07 joink split the sensor?
14:07 norbi so 2 of those cameras
14:07 norbi joink: ?
14:07 norbi with a knife?:P
14:07 joink nah, send two optical signals to it divided by something in the center :P
14:07 joink but yeah. get two
14:11 norbi joink: ok, i understand
14:11 norbi thanks
14:11 norbi this will be interesting:D
14:12 norbi i feel the vibration:P
14:12 norbi im thinking what mcu would need this
14:12 norbi at what speed
14:12 norbi maybe 80Mhz would be enough
14:12 norbi for getting the image and processing it
14:14 amee2k mmh vibration is a good point
14:14 amee2k depending on the suspension of your rc car you'll need to dampen vibrations or the image will get all blurry
14:16 amee2k and depending on what speed you're looking for you need fairly high framerate
14:16 norbi slowly
14:17 amee2k an accelerometer would help though because it is much faster than determining velocity and acceleration visually
14:17 norbi the thing will be that to prove the concept
14:17 norbi i do have the adxl345
14:17 norbi you know that i played with it last week
14:17 thylanefr lol. this conversation
14:17 amee2k http://c29494.r94.cf1.rackcdn.com/wp-content/uploads/2011/01/dale.jpg ?
14:17 amee2k :)
14:18 norbi thylanefr: ?:)
14:21 thylanefr style points on the slalom
14:22 amee2k well, that too
14:22 amee2k but extra style points on drifting *nod*
14:23 norbi :))
14:23 amee2k and, erm, extra extra style points for drifting *and* using the RC model of a panda-colored tureno gt-apex
14:23 norbi :))
14:23 ee2k idly wonders who got that reference
15:04 aggrav8d I've got a 24v12a power supply for stepper motors and a 5v power supply for the control logic. the 5v power comes from an arduino. the control board docs say I need a common ground. how can i wire a common ground without frying the arduino?
15:06 Tom_itx you won't
15:06 Tom_itx just run a gnd wire to all
15:09 Tom_itx +
15:09 Tom_itx oops
15:30 aggrav8d oh yeah? There won't be feedback or anything?
15:30 aggrav8d so i can just run a wire from arduino GND to 24v PS GND and call it a day?
15:31 Tom_itx yup
15:31 Tom_itx certain circumstances it's best to do what is known as a star ground configuration
15:32 Tom_itx where all grounds come together at one point
15:32 Tom_itx instead of gnd to gnd to gnd.....
15:35 aggrav8d that'd be nice but I have three power transformers. Not sure how that would work. :T
15:35 aggrav8d 12v power for the DC motors, 24v for the steppers, and 5v for the logic.
16:30 norbi i dont really understand this problem with gnd
16:30 norbi this part is a bit vague for me
16:31 Tom_itx helps eliminate noise
19:06 aggrav8d hm.
19:06 aggrav8d so I wired it all up with MM160 stepper drivers from Solarbotics (http://www.soc-robotics.com/pdfs/MM160%20Technical%20Reference%20Manual%20R1.0.pdf) and I find I'm getting thermal shutoff within a few seconds of power on, even with forced air cooling.
19:07 aggrav8d the two 3a steppers are plugged into a 12.5a power supply. There's going to be a third when I've tested the first two. What's causing all the heat?
19:08 aggrav8d I thought at long as the volts matched it would be ok.
20:33 Tom_itx aggrav8d, you need some sort of current limit. stepper drivers are typically chopper drive