#robotics Logs
Nov 15 2017
#robotics Calendar
12:03 AM rue_shop3: and I know I'm in danger when I say to rue "but you dont need that" and the only reply and "aagaagaggagagagaaaa" and a mess of droool
01:47 AM logstaren: im trying to watch my spending
01:47 AM logstaren: i dont make much
01:48 AM logstaren: will barely have a few Ks saved after paying for taxes (gotta send the money myself as a contractor), after working for a year
01:48 AM logstaren: this deal was quite bad
01:48 AM logstaren: but then again, first real job..
07:42 AM robotustra: Tom_L, this is a test plate, but I'll use it as a rebar in my lathe. This surfece will not be used.
07:44 AM robotustra: Tom_L, I use this scraper https://www.ebay.com/itm/Utility-Knives-The-Original-SuperScraper-SS-5-Double-Edge-Carbide-Scraper-/302265447852
07:49 AM Ultrasauce: isnt that going to gouge the hell out of the print bed?
07:51 AM robotustra: I don't understand your English, sorry
07:51 AM Ultrasauce: if the scraper is harder than the print surface, won't it damage it?
07:53 AM Ultrasauce: nevermind, failed reading comprehension
07:53 AM robotustra: what printing surfece?
07:53 AM robotustra: what is print surfece?
07:54 AM robotustra: this scraper is harder than surface material, that's why it SCRAPES and remove some material
07:54 AM robotustra: it's tungsten carbide and it removes steel particles
07:54 AM Ultrasauce: yeah i get it now, thought the topic at hand was 3d printing
08:03 AM AchiestDragon: should realy finish off my printer https://www.youtube.com/watch?v=ZUKkUsrWyn0 i replacwd the nozzle with a commertial one that blocked after printing the first layer then managed to melt the heated bed relay and burn out its psu
08:04 AM AchiestDragon: replacwd /replaced
08:04 AM AchiestDragon: its been sat untuched for 4 years now gathering dust and rust
08:06 AM robotustra: AchiestDragon, I think in the #reprap channel ppl can help you more
08:07 AM robotustra: or you don't need any help with that :)
08:07 AM AchiestDragon: :)
08:09 AM robotustra: it's more like cnc
08:09 AM robotustra: huge motors
08:10 AM robotustra: industrial grade
08:10 AM robotustra: you have to finish it
08:11 AM Ultrasauce: motivation is definitely the hard part
08:11 AM Ultrasauce: i finally got around to replacing my awful mendel
08:11 AM robotustra: do you know what is positive feedback?
08:12 AM AchiestDragon: yea , and drivers , i recycled the parts from my old cnc router , since i had them laying arround , its mega fast , looks like it wobles all over but thats the big rubber shock mounts that stop it over skipping when the motion stops , they take out all the inerture
08:12 AM AchiestDragon: and has to slow to a snales pace when extruding
08:13 AM Ultrasauce: newer controller firmware will let you cap the acceleration also to prevent that sort of issue
08:15 AM robotustra: sounds like advertisement of _NEWER_CONTROLLER_
08:16 AM AchiestDragon: yea it could do with a more up to date 32bit one rather than the 8bit atmega
08:16 AM Ultrasauce: yes i am a total shill for anything not arduino based
08:16 AM Ultrasauce: you got me robotustra
08:17 AM AchiestDragon: actualy had been contemplating https://www.ebay.co.uk/itm/TEVO-Little-Monster-Delta-3D-Printer-High-Precision-Large-Building-Size-DIY/322849056651?ssPageName=STRK%3AMEBIDX%3AIT&_trksid=p2060353.m1438.l2649
08:17 AM AchiestDragon: to build robot parts on :)
08:18 AM Ultrasauce: that's the one i have, highly recommend
08:18 AM Ultrasauce: you can get a better price from a direct vendor
08:19 AM robotustra: when I hear about "super precision" printers I'm lauphing
08:19 AM AchiestDragon: im dew some backpay , so depends on how much i have left over when it arives
08:19 AM robotustra: especially tall delta printers
08:20 AM Ultrasauce: in practice 0.3mm feature size is attainable on that one
08:20 AM robotustra: the precision of deposition is ~ 1/2 of nozzle diameter
08:21 AM robotustra: if all mechanics give you lashbacks less than 0.1 mm
08:22 AM robotustra: if it's tall delta - you'll get more because of lack of rigidity
08:22 AM AchiestDragon: the one i was building was designed with a 300mm,600mm,300mm print aria , the tveo is not far of the same
08:22 AM robotustra: theoretically delta can be very good
08:22 AM Ultrasauce: it's reasonably rigid, big aluminum extrusions. not perfect of course
08:23 AM robotustra: to make delta rigid it should be closed hex prism
08:23 AM robotustra: and put magnet's and balls in every joint
08:24 AM robotustra: even in this case - plastics shrink with the temperature
08:24 AM Ultrasauce: yeah perfect precision is impossible
08:24 AM robotustra: and it gives substantional eroor
08:25 AM robotustra: error
08:25 AM AchiestDragon: yea , the only issue is i would have to sort a guard cage for it as my cat would attack it while its printing
08:25 AM Ultrasauce: maybe once :)
08:25 AM Ultrasauce: it only took one burn from the hotend to make me stay the hell away
08:26 AM robotustra: I decided not to use printed parts for critical places for my robot
08:26 AM robotustra: will use polycarbonate, aluminum, steel
08:27 AM robotustra: PLA stuff only for non critical boxes and cpacers
08:27 AM robotustra: I don't trust 3d printed parts can be durable
08:28 AM robotustra: I mean plastic 3d printed parts
08:28 AM robotustra: which are printed on home made 3dprinter :)
08:31 AM AchiestDragon: for experimentation should not be an issue , but for long term use then they would be more suseptable to breaking and wair
08:34 AM AchiestDragon: anyway , main question i was going to ask was does anyone here use PCL ( point clould libraries ) with sterio webcam here inplace of ultrasonics for object avoidance
08:37 AM AchiestDragon: was reading this https://www.researchgate.net/publication/272823931_Low_Cost_Stereo_System_for_Imaging_and_3D_Reconstruction_of_Underwater_Organisms
08:37 AM AchiestDragon: seems its possible to do on a rpi3
08:38 AM Ultrasauce: doing SLAM is on my list but haven't touched it yet
08:41 AM SpeedEvil: https://www.ebay.co.uk/itm/Mini-3D-Photograph-Stereoscopic-Camera-Lens-w-Clip-For-iPhone-Smart-Phone/262891018187 I have noted before.
08:41 AM SpeedEvil: This eliminates the issue with synchronisation of the cameras
08:42 AM Ultrasauce: https://github.com/HKUST-Aerial-Robotics/VINS-Mono/blob/master/support_files/paper/tro_technical_report.pdf
08:42 AM Ultrasauce: doesn't even need stereo
08:43 AM Ultrasauce: https://www.youtube.com/playlist?list=PLpUPoM7Rgzi_7YWn14Va2FODh7LzADBSm
08:52 AM rue_mohr: robotustra, I'v been surprised, good prints at 40% or higher infill can be amazingly strong and hold up really well
08:53 AM rue_mohr: me and a guy made a curtin rod holder, and when we tried to do a fail test, we gave up at about 15kg cause we were so far over what the curtin weighed
08:54 AM veverak: wow
08:54 AM Ultrasauce: yeah the issue for me is the slicer can do dumb stuff that compromises the strength
08:55 AM rue_mohr: I'm using slic3r and cant say I'v has issues
08:56 AM robotustra: I use slic3r also, no problem what so ever
09:04 AM rue_mohr: I have more problems with the plastic feed jamming and leaving slits in the prints
09:07 AM rue_mohr: SpeedEvil, sweeeet, thanks, I was trying to design a lens like that
09:07 AM rue_mohr: $3 DONE!
09:07 AM robotustra: I never had feed problem on my romax
09:07 AM SpeedEvil: rue_mohr: sometimes mass produced crap works fine.
09:07 AM rue_mohr: dosn't help when you get a pencil tangled up in your spool holder
09:08 AM robotustra: I see that rue_mohr try to do everything for negative price
09:08 AM rue_mohr: or if the 1.75mm has a 3mm lump in it
09:09 AM rue_mohr: tends to cause jamming
09:12 AM AchiestDragon: main isue i had last time i tried doing a stereo webcam setup , i just 2 usb wecams placed a known distance appart ended up beeing dew to capture program issues
09:12 AM rue_mohr: yea
09:13 AM rue_mohr: I wanted to split one camera casue otherwise you have sync or latency issues
09:13 AM rue_mohr: digital is horrid for latency
09:13 AM AchiestDragon: in that after trying a number of methods the frame rate that should of been about 30fps was reduced by software from the driver down to between 2 to 8 fps
09:13 AM rue_mohr: there are something like 30 buffered frames in even a simple digital stream
09:13 AM robotustra: I have 2 cams working at 17 fps
09:14 AM rue_mohr: nobody seems to know how to properly stream data
09:14 AM robotustra: don't use shitty biggle bone
09:14 AM rue_mohr: they all decide its ok to buffer up a frame before starting to send it out
09:14 AM robotustra: my cameras give unpacked frames
09:15 AM robotustra: at 30 fps
09:15 AM rue_mohr: yea, but what does all the software handling it do
09:15 AM robotustra: software does nothing yet
09:15 AM AchiestDragon: i got as far as finding i needed to mod the source for wxcam to increase the default frame rate but left it for other projects
09:15 AM robotustra: I just store images at this pace
09:16 AM rue_mohr: so your frame buffering
09:16 AM Ultrasauce: using separate usb root hubs is key for multiple webcams i think
09:16 AM robotustra: and frame rate depends on the light strongly
09:16 AM AchiestDragon: lol
09:16 AM robotustra: in the dark it drops to 16 fps
09:16 AM Ultrasauce: usb2 bandwidth is quite limited and shared between all devices on the same hub
09:16 AM rue_mohr: linux recently put bandwidth checking on usb
09:17 AM Ultrasauce: and isochronous transfers can only allocated 80% of the available bandwidth
09:17 AM rue_mohr: so I cant use my audio card cause there are two unused serial converters on the same hub
09:17 AM robotustra: Ultrasauce, even if you have 2 different hubs - the bottleneck is v4l driver
09:18 AM Ultrasauce: i don't think so, i've done dual 640x480@120fps
09:18 AM robotustra: at least I have this issue on my linux system
09:19 AM AchiestDragon: had 2 seprate usb hub controler cards seperate pci ones runing in the hp dl740 server (8 xeon cpus) , the other reason was that server was only 32 bit and pcl swiched to 64 bit builds , a server upgrade at the time was not within budget
09:19 AM Ultrasauce: there is certainly a lot on the software side that can go wrong also
09:19 AM robotustra: Ultrasauce, what gives your cams?
09:20 AM robotustra: my cam can't give a stream, it gives unpacked frames
09:20 AM AchiestDragon: and was only getting 8fps from any cam with 30 fps for the first couple of seconds after initalising them
09:20 AM rue_mohr: so, a sub-50c dac will require me to build my own R-2R ladders
09:20 AM robotustra: 900k per frame 640x480x24
09:20 AM Ultrasauce: just the usual yuyv frames
09:21 AM robotustra: Ultrasauce, is it USB3?
09:21 AM Ultrasauce: no
09:21 AM rue_mohr: 27M/s
09:21 AM rue_mohr: 30fps assumed
09:21 AM robotustra: Ultrasauce, explain me how you did it
09:22 AM Ultrasauce: dedicated pci usb card for each camera, vlc to capture
09:22 AM rue_mohr: 37sec/G
09:22 AM robotustra: Ultrasauce, what OS?
09:23 AM Ultrasauce: didn't achieve low latency with that configuration though, had to sacrifice throughput to get that
09:23 AM Ultrasauce: linux
09:23 AM AchiestDragon: yea the hw is not the issue the driver throtles the video , i got as far as descovering theres a seting in there that reduces the fps to conserve usb bandwidth , but not as far as solving how to change it
09:23 AM robotustra: I tried to do it with v4l driver directly
09:23 AM rue_mohr: for now, I'm skipping the digital stage
09:23 AM AchiestDragon: or at least thats how it was about 4 years back when i tryed
09:23 AM Ultrasauce: these days i'm working with SDI video and hardware h265 encoders
09:23 AM rue_mohr: I want to keep the stereo vision one analog stream
09:24 AM rue_mohr: hence the optical splitter
09:24 AM rue_mohr: eee gad I have to go to work
09:24 AM Ultrasauce: cant really get the latency below ~100ms but there sure are a lot of pixels :)
09:24 AM robotustra: actually even dual 17 fps is good enough to make a robot vision
09:25 AM robotustra: because I think I'll spend more time to process frames
09:25 AM robotustra: that 50 ms per frame
09:25 AM robotustra: than*
09:26 AM robotustra: to capture image is the very first and the simplest step
09:26 AM robotustra: all fun starts after that
09:28 AM robotustra: Ultrasauce, do you process your images?
09:29 AM Ultrasauce: yes but not for CV applications
09:30 AM Ultrasauce: the thing i'm working on now is for film&television production
09:30 AM robotustra: Ultrasauce, so you do 2D image processing?
09:31 AM robotustra: or 3d stuff also?
09:31 AM Ultrasauce: just 2d at the moment. scene mapping and stuff further down the roadmap
09:31 AM robotustra: I want to make stereo vision for my robot
09:32 AM robotustra: but it's still in development
09:32 AM Ultrasauce: i recommend the tegra x1
09:32 AM robotustra: I'm planning to use up-board
09:33 AM robotustra: is there any single boards with this processir?
09:34 AM Ultrasauce: jetson tx1
09:34 AM Ultrasauce: and then you can use a real video interface instead of usb, and cuda video processing
09:36 AM robotustra: I see already a couple of problems
09:36 AM robotustra: 1) cooling
09:36 AM robotustra: 2) power consumption
09:36 AM Ultrasauce: <10W running all out
09:36 AM robotustra: it's A LOT for mobile robot
09:36 AM Ultrasauce: it will use less doing the same processing as that intel board, i guarantee it
09:37 AM robotustra: I'm pretty sure it will work ok
09:37 AM robotustra: but software should be done anyway
09:39 AM robotustra: tx1 is too expensive for now
09:39 AM AchiestDragon: i was looking at setting up a local server accesable over wifi to do the main processing that way i could have a number of small robots sharing the same mapping info and higher cpu speeds not posible onboard
09:40 AM Ultrasauce: realtime video over wifi really sucks
09:40 AM Ultrasauce: i say that as someone basing his livelihood on that very thing
09:40 AM robotustra: wifi sucks
09:44 AM AchiestDragon: so does usb
09:50 AM AchiestDragon: althoug still think there maybe some scope in using the 2 rpi's with cams (not usb ) and a onbord lan using them to process in paralell , with pi3's a bit of streem allignment code , maybe a good cheap solution and power consumption/ and size is not that bad
09:53 AM AchiestDragon: yea $599 even a small cluster of 8 rpi's would be more cost effective
09:55 AM Ultrasauce: if you really wanna do realtime cv on the pi you are more than welcome
09:59 AM AchiestDragon: you should read the pdf i posted from the guy that was doing it , it covers quite a bit of what he had to do to get the performance good enough to mange it on the pi , something that also improves the perfromance of the PCL libs on other processes also
10:02 AM Ultrasauce: you should read the pdf and note that it's all offline processing and thus unsuitable for mobile robotics
10:02 AM robotustra: true
10:03 AM robotustra: real time impose a lot of restrictions
10:03 AM robotustra: real time processing*
10:03 AM AchiestDragon: hence why i was looking at steaming to a local server for it
10:04 AM robotustra: humans can capture like 25 fps
10:04 AM robotustra: if robots are slower, why should they do 120 fps?
10:04 AM robotustra: just because they CAN?
10:04 AM AchiestDragon: depends on the app for the robot
10:05 AM robotustra: common household tasks
10:05 AM robotustra: pick up things from the floor, range them
10:05 AM robotustra: load laundry and washing machine
10:06 AM robotustra: clean the floors
10:06 AM AchiestDragon: ie for something like cv on say https://www.youtube.com/watch?v=XF4jExjWl50 micromouse then 120fps would be good but not what there using
10:07 AM robotustra: and I would prefer not to use 3rd party libraries like OpenCV
10:08 AM robotustra: to not depend on their bugs and updates
10:08 AM SpeedEvil: yeah.
10:08 AM SpeedEvil: reimplementing openCV is a good idea.
10:09 AM SpeedEvil: That will certainly have less bugs and less effort invested in a robust solution.
10:10 AM robotustra: actually I'll need like 5% of what they have, and add 50% new things they don't have
10:10 AM robotustra: I know how to multiply matrixes
10:11 AM robotustra: and calculate cosines and sines
10:13 AM robotustra: for instance OpenCV does not have support of stereovision for moving eyes
10:13 AM deshipu: I can never understand that 'I can't use tyyhis library because it does more than I need' excuse. Just admit you are too lazy to read the docs.
10:13 AM robotustra: deshipu, unfortunately I did read it
10:13 AM robotustra: now I can't forget it back :)
10:13 AM veverak: deshipu: it is proper excuse
10:13 AM veverak: in some cases
10:13 AM veverak: :)
10:13 AM AchiestDragon: :)
10:14 AM deshipu: also, there is this 'I don't know how to tackle the hard part of my problem, so I will focus on reimplementing the trivial parts' thing.
10:14 AM robotustra: my code in C takes like 100k to do everything I need, why I should install shit pile of megabites of OpenCV?
10:14 AM veverak: deshipu: using external library, is overhead by itself
10:14 AM deshipu: but of course it's all legit if you do it for fun
10:14 AM veverak: deshipu: and you can find out that a lot of libraries are general solvers, so they have general API
10:14 AM veverak: that may not be ideal for your use case
10:15 AM robotustra: deshipu, actually I'm a guy who know how to tackle hard problems
10:15 AM deshipu: great
10:15 AM veverak: so, if I need something that is simple to implemer to implement, than to manage external dependency and having to adhere to the api of library, I do it
10:15 AM AchiestDragon: robotustra: yea have to say for pro use its prob not the best , but for hobby use it just seems over complex
10:16 AM veverak: deshipu: but hey, discussion about when it is proper case can be never-ending
10:16 AM robotustra: I observe the modern aproaches in "doing things": 1) quick and dirty 2) slowly but surely and robust
10:17 AM deshipu: veverak: personally I like to cobble together a prototyppe with as little work as possible, and if it's too slow, then optimize
10:17 AM robotustra: I prefer the 2 way
10:17 AM edmoore: it's educational to do your own cvis algorithms
10:17 AM robotustra: quick and dirty is foor for prototyping
10:17 AM deshipu: saves me a lot of work when my approach proves to not work
10:17 AM edmoore: but it's unlikely that knowing the very basics of linear algebra and trigonometry will allow you to make something more robust than openCV
10:17 AM veverak: deshipu: I do prototypes in python because of that, "python script, import what you need and try"
10:17 AM robotustra: good for*
10:18 AM veverak: is simplerand most effective
10:18 AM veverak: however
10:18 AM veverak: deshipu: for something that should work in lognterm, I do not use anything that is not an package
10:18 AM veverak: in either ROS or distribution
10:18 AM veverak: which drops a lot of github repositories
10:18 AM veverak: :)
10:18 AM robotustra: dependencies are evil
10:18 AM veverak: yeah
10:18 AM deshipu: veverak: I got into the whole walking robot thing because other interesting behaviors of robots were too hard for me
10:19 AM edmoore: i used to work in a real-time computer vision research lab, especially on fast feature extraction and tracking, and making it both fast and robust is a research problem still.
10:19 AM veverak: dependency in proper project means, that you have to watch other project, so they do not screw something up for you
10:19 AM veverak: that overhead should compensate what it provides you
10:19 AM deshipu: robotustra: dependencies are great, they save you loads of time in maintenance
10:20 AM robotustra: deshipu, I beleive only in libc dependency
10:20 AM robotustra: :)
10:20 AM veverak: deshipu: and there is always the problem of design and API
10:20 AM robotustra: all the rest I can do myself
10:20 AM deshipu: projects built on beliefs alone don't usually work too well
10:21 AM robotustra: deshipu, as you wish
10:21 AM veverak: deshipu: the libraries have they waye of providing API, sometimes, that means that it's better to create something custom with API that fits your needs
10:21 AM robotustra: how is your walking creature?
10:21 AM edmoore: i'm all for this if it's just for a hobby and you want to play around and enjoy the journey
10:21 AM deshipu: robotustra: it's walking
10:22 AM robotustra: deshipu, walking good?
10:22 AM deshipu: right, if it's just fun, any approach is legit
10:22 AM deshipu: robotustra: yes, thank you
10:22 AM robotustra: I got one thing recently
10:23 AM deshipu: robotustra: still not doing anything else, though
10:23 AM edmoore: robotustra: if you want to start playing, maybe try and implement something like the SIFT algorithm in c for yourself
10:23 AM edmoore: that's a good basic feature detection algorithm
10:23 AM edmoore: it's a bit too slow for real-time stuff though, but it's a good starting point to get into the headspace
10:23 AM AchiestDragon: depends how slow the robot is
10:23 AM robotustra: edmoore, what is SIFT?
10:24 AM branjb: similar to surf
10:25 AM edmoore: robotustra:
10:25 AM edmoore: 16:23 <edmoore> that's a good basic feature detection algorithm
10:25 AM robotustra: "a bit too slow" is not acceptable for me. if I have 17 fps, it means I have about 50 ms to process 2 frames, and all my code should be able to do processing in time
10:25 AM deshipu: make it work first, make it fast later
10:25 AM robotustra: and I don't need basics, I need it complex 3D meshed solids
10:26 AM edmoore: robotustra: yes but it's also a good start
10:26 AM edmoore: so my point is, implement it in C
10:26 AM robotustra: deshipu, I don't accept to postpone "make it fast later"
10:26 AM robotustra: edmoore, it is implemented in C
10:27 AM edmoore: but you have just been saying you want to implement it yourself
10:27 AM edmoore: so _you_ should implement it in C
10:27 AM edmoore: make sure you understand the maths, which is quite simple compared to a lot of cvis algorithms
10:27 AM deshipu: robotustra: well, then you won't get it working
10:28 AM robotustra: edmoore, I want to give vision to this guy https://www.youtube.com/watch?v=om_hhY21Tbg
10:29 AM robotustra: edmoore, "make sure you understand the maths" - yes I understand what I'm doing
10:30 AM deshipu: what was the question then?
10:30 AM robotustra: this guy has 2x 640x480 @17 fps
10:30 AM deshipu: sorry, I forgot
10:30 AM edmoore: right, so you will have no problem understanding sift
10:30 AM edmoore: then once you do we can talk about some more advanced techniques that use some machine learning to help
10:30 AM robotustra: edmoore, I'm alredy far beyond sift
10:30 AM AchiestDragon: https://lh3.googleusercontent.com/-gg-o8W8fSoE/WgxrQiOoytI/AAAAAAAABfo/nDmul3QqJ8sr-KTXocjSdqznalNcCE8LgCL0BGAs/w530-d-h353-n-rw/tinybug-oct07.jpg
10:31 AM robotustra: AchiestDragon, what is this?
10:32 AM AchiestDragon: pic of my last robot , link work ?
10:32 AM edmoore: robotustra: ok cool, so what algorithms are you looking at for this then?
10:32 AM edmoore: you can link to papers if you want
10:33 AM robotustra: AchiestDragon, it downloads something which is cannot be open
10:33 AM AchiestDragon: that was recycled and is now the also not yet functional 3d printer , in attempt n to get a fucntional robot working ,, like the link apparantly
10:34 AM robotustra: edmoore, I'm designing an algorythm to detect common points on 2 different pictures
10:34 AM robotustra: for the system with memory
10:34 AM edmoore: right, and what techniques are you using for this
10:34 AM robotustra: a set of techniques
10:35 AM edmoore: for example
10:35 AM edmoore: you can surely list a few, this is a well-researched area of computer vision and pattern recognition
10:36 AM robotustra: edmoore, I'm not planning to use NN to detect something on the images. but I do use kind of pattern matching
10:36 AM edmoore: what kind of pattern matching?
10:37 AM robotustra: of small samples taken from other frame
10:37 AM robotustra: I use diff
10:37 AM robotustra: and slide the pattern
10:38 AM edmoore: and how to you account for changes in lighting, subject orientation, noise etc?
10:38 AM robotustra: and looking for extremums
10:38 AM robotustra: images are equalized
10:38 AM robotustra: of cause
10:38 AM robotustra: I have 2 cameras, looking approximately the same directin, so the light conditions ate about the same
10:39 AM edmoore: and how do you account for changes in subject orientation and sensor noise?
10:40 AM robotustra: that's the good question, but as soon as I got from my experiments, if to take small area of the sample - I have no problem with orientation of the subject
10:41 AM AchiestDragon: this link better https://drive.google.com/file/d/1gYI7hEC-aZF7QqyaGCsIQG06wfsvGyng/view?usp=sharing
10:41 AM robotustra: sensor noise is negligible in my case
10:42 AM edmoore: robotustra: i'm either misunderstanding you or ewhat you're describing is extremely naive and unrobust
10:42 AM edmoore: just convolving a segment of one image with another
10:42 AM robotustra: AchiestDragon, does this thing move
10:43 AM edmoore: anyway, if you want to read the paper that my old principal investigator in my lab wrote, which is still about the best way of doing feature extraction for real-time computer vision applications, it's here: https://www.edwardrosten.com/work/rosten_2006_machine.pdf
10:43 AM robotustra: edmoore,one does not exclude another.
10:44 AM AchiestDragon: it did , it managed 0.1mph and is power consumption was 1kw/h ,and could manage all of 6 foot before needing recharging that had to be done in stages totaling 48 hrs to charge
10:44 AM robotustra: edmoore, you do exactly the same
10:45 AM robotustra: you do the summation of the interposed patterns
10:45 AM edmoore: it's not really a summation though
10:45 AM edmoore: your brain does a lot more inference work than that
10:45 AM robotustra: really? what does NN do in your case?
10:46 AM robotustra: weghted summation
10:46 AM edmoore: you said you weren't using NNs
10:46 AM robotustra: and as I told, I want to avoid usnig NN
10:47 AM edmoore: and don't be fooled into thinking that because we call a graph-based optimisation network a 'neural network' that it's a perfect model of what the brain does
10:47 AM robotustra: yes I don't use it explicitly, but for some operation I need somewhat similar to find extremoms
10:47 AM AchiestDragon: the only other interesting point was it used cheap sat dish positioning jacks (18 of them ) as they where half the cost of metal geared r/c servos
10:48 AM robotustra: edmoore, I know what are you talking about, but when you go to small robotics with small CPU, all NN stuff start to suck heavily, and I suck with them too
10:49 AM edmoore: yes
10:49 AM edmoore: i agree
10:49 AM robotustra: it's too CPU and GPU consuming stuff
10:49 AM Ultrasauce: if only there were a gpgpu option
10:49 AM edmoore: i never used NNs when i was in cvis research
10:49 AM edmoore: it's only exploded in popularity since about 2010, which is when i left academia
10:50 AM robotustra: that's why I'm trying to make simplified algorythm, even if they will be not so robust from the beginning
10:50 AM edmoore: all the stuff i'm talking to you about, especially stuff that needs to be fast, does not involve NNs
10:50 AM robotustra: and that will run on small single board computers
10:50 AM edmoore: if you think i'm advocating NNs then you are not reading me properly
10:50 AM robotustra: edmoore, ok
10:51 AM edmoore: the paper i gave you is specifically designed for being good on cpu resources
10:51 AM edmoore: it was and still is used a lot in mobile robotics
10:51 AM robotustra: the article describes how to detect common points from 2 very distinct camera
10:51 AM edmoore: a lot of the early smartphone augmented reality apps used it, that was back in 2010 when smartphones were very anaemic compared to modern ones
10:51 AM robotustra: distant*
10:52 AM AchiestDragon: the other point in cv on small robots is reducing the res and only dealing with close range objects , you dont need to reolve a data set based on the whole image and range
10:52 AM AchiestDragon: reolve / resolve
10:52 AM robotustra: in the case of binocular vision everything is getting easier, because the picture you get is quite similar
10:53 AM robotustra: AchiestDragon, as I told, do you know how much OpenCV takes on your hard drive?
10:53 AM edmoore: robotustra: that's not really true at all
10:53 AM edmoore: especially at close range
10:54 AM robotustra: edmoore, yes but no
10:54 AM edmoore: you either have to have some invarient representation (i.e. a system agnostic to the angle of view) or you don't. If it's a different camera in a different position with a different sensor then you really do
10:54 AM robotustra: edmoore, as I told, I'm not doing a bullet proof algorythm to handle all cases
10:55 AM edmoore: yes that much is obvious
10:56 AM robotustra: AchiestDragon, did you build your robot?
10:57 AM robotustra: AchiestDragon, I mean is it your hexapod?
10:57 AM AchiestDragon: the other point with cpu and openCV is the fact its written to cover large robots and hi res processing somthing that is not needed on a tiny "toy" sized robot
10:57 AM AchiestDragon: yea
10:58 AM robotustra: edmoore, my goal is to make a system with memory which can create a 3d meshed map of objects, by finding and adjusting points in 3d space, with 2 cheap webcams and at 15-20 fps
10:59 AM robotustra: and make it working on a small CPU
10:59 AM robotustra: like intel atom
10:59 AM AchiestDragon: was my first attempt at a hexipod , i dont have pics of the second but i had ran out of gates in the fpga i was using on that
11:00 AM robotustra: of cause I partially use algorythms from OpenCV and other papers but the are too heavy for my case
11:00 AM robotustra: AchiestDragon, is it walking?
11:00 AM AchiestDragon: not shure if to try another hexipod or build a 2 legged walking design
11:01 AM robotustra: edmoore, for instance, I can do fast edge detect by simple subtracting of 2 subsequent frames
11:01 AM robotustra: just because of noise on the sensor
11:01 AM robotustra: as simple as that
11:01 AM AchiestDragon: nolonger , that large hexipod was striped and the parts reused on other projects and the other was scraped
11:02 AM edmoore: robotustra: i'm not sure they're too heavy
11:02 AM robotustra: and it's good enough for my goals. Making edge detect with gaussians and bluring takes much more time
11:02 AM edmoore: have a look at this from 8 years ago, done with a single webcam https://www.youtube.com/watch?v=1UCiSb3D37A
11:03 AM edmoore: that uses the fast corner detection algorithm i linked to in the paper
11:03 AM edmoore: we had it running on a phone
11:03 AM AchiestDragon: i do it for fun , so , whatever if it keeps me busy
11:04 AM robotustra: edmoore, single camera is not the same as 2 cameras
11:05 AM edmoore: i know that but the principles are about the same
11:05 AM robotustra: edmoore, but it's a good demo thow
11:05 AM edmoore: 1 view + time is similar to n views
11:06 AM edmoore: the reconstruction logic isn't too worried about the units of the dimensional space
11:06 AM AchiestDragon: edmoore: kool
11:06 AM robotustra: yes, the same, I will use triangulation to avoid camera tistortions for instance
11:06 AM edmoore: right
11:06 AM edmoore: so why correct me
11:07 AM edmoore: do you mind if I ask how old you are
11:07 AM robotustra: 43
11:07 AM robotustra: I was a prof in Uni for 10 years
11:07 AM robotustra: in physics
11:07 AM robotustra: in computational physics
11:08 AM edmoore: when did you leave?
11:08 AM robotustra: because I moved to canada and went to embedded programming
11:08 AM edmoore: when i said robotustra
11:08 AM edmoore: not why
11:09 AM veverak: interesting
11:09 AM robotustra: 2010
11:09 AM edmoore: so you left when you were 36
11:10 AM edmoore: so you became a professor of computation physics at 26
11:10 AM edmoore: wow
11:10 AM edmoore: well you're a genius and you certainly don't need my help
11:10 AM edmoore: good luck
11:10 AM robotustra: yep
11:10 AM robotustra: I know linear algebra and all that stuff
11:10 AM edmoore: i bet you do
11:10 AM robotustra: I did some Shrodinger equation solving
11:11 AM edmoore: what self respecting professor of computational physics wouldn't
11:11 AM robotustra: huge matrix diagonalisation
11:12 AM robotustra: who told you that I'm self respecting?
11:12 AM robotustra: :)
11:12 AM edmoore: and why did you diagonalise the matrix?
11:13 AM robotustra: because the numerical solution of Shrodinger equation need it
11:14 AM edmoore: sure
11:14 AM AchiestDragon: Shrodinger , was he that guy with the cat in a box experiment , you know the one where the cat managed to develop a quantum transfer drive and escape to someware they dont put it in a box
11:14 AM edmoore: what does diagonalisation give you
11:14 AM edmoore: i am familiar with schrodinger
11:14 AM edmoore: i'm just seeing if you're actually familiar with linear algebra
11:14 AM Snert_: What happens when Pavlov's dog sees Schrodingers cat?
11:15 AM edmoore: you get someone pretending to be something they're not on irc
11:15 AM robotustra: edmoore, do you know what is LAPW?
11:15 AM edmoore: and then they disappear when you try to test them
11:15 AM robotustra: or LMTO?
11:16 AM edmoore: no i'm afraid not
11:17 AM AchiestDragon: lmto = laugh my tits of
11:17 AM robotustra: edmoore, do you know density functional theory?
11:17 AM robotustra: LMTO = Linear Muffin-Tin orbital
11:18 AM robotustra: LAPW = Linear augmented plane waves
11:18 AM robotustra: DFT = dencity functional theory
11:18 AM AchiestDragon: depends on the acronim shortand your using
11:18 AM edmoore: i'm a rubbish physicist i'm afraid, and have never really done the maths required for any QM stuff
11:19 AM edmoore: the best i have done was start sussman's book on differential geometry and not finish it because of problems with the scheme interpreter. not much to show for it
11:19 AM robotustra: edmoore, long story short, you represent the wave function in some basis, and solve shrodinger equation in a matrix form
11:20 AM robotustra: and there you'll get a lot of math physics and linear algebra
11:21 AM robotustra: anyway, I played math on PCs a lot and know how much usually matrix stuff takes
11:23 AM edmoore: i only asked you about diagonalisation to see if you'd tell me about eigenvalues, but if you actually are explaining bits of differential geometry i don't really understand then i'm more inclined to believe you
11:27 AM robotustra: edmoore, I'm here not to prove that I'm a prof of physics, but yes, the diagonalisation is used to determine eignvalues
11:28 AM robotustra: and I did it for solids for the set of K vectors in reciprocal space
11:28 AM robotustra: to build band structure
11:29 AM robotustra: edmoore, really, I don't care you trust me or not
11:29 AM edmoore: well, i just recommend then vacuuming up a few papers of fast robust cvis
11:29 AM edmoore: you should find them no trouble
11:31 AM edmoore: do you have a copy of 'Pattern Recognition and Machine Learning' by Bishop?
11:31 AM edmoore: it's a very hand reference
11:31 AM robotustra: probably I have it somewere on hard drive
11:32 AM edmoore: i usually keep it around if i'm reading any papers on any of those topics
11:32 AM AchiestDragon: the only Cv stuff i have done profesionaly was i used to bench repair ocr scanners , many years back
11:33 AM robotustra: yes, I looked throw this book a while ago
11:35 AM robotustra: but real men invents croaches and bikes
11:36 AM robotustra: NN is just another representation of regression
11:40 AM edmoore: you've lost me again
11:40 AM robotustra: edmoore, no, you just lost yourself :)
11:41 AM robotustra: I do kind of gradient descending
11:42 AM robotustra: and yes, the math is similar to the techinque which is used inside NN
11:42 AM robotustra: but I cut corners trying to make it faster
11:42 AM robotustra: it's more like functional minimisation for me
11:44 AM robotustra: but anyway, thanks for your advise
11:46 AM robotustra: edmoore, you should know than I'm not a math prof, I'm a physicist. Physics deals with approximate stuff, not strict stuff :)
11:48 AM edmoore: sure, i am an engineer
11:48 AM edmoore: so likewise
11:49 AM edmoore: tho programming is often a good test of mathematical understanding an ambiguity of notation
11:50 AM edmoore: you have probably found than in QM
11:50 AM edmoore: an equation looks ok until you actually try to program it and realise some of it is a bit ambiguous
11:50 AM edmoore: what exactly is the pde with respect to, etc
11:53 AM robotustra: yeah, to program formulas is the pain in the ass
11:54 AM robotustra: I spent about 1 year to make my technique working properly
11:54 AM robotustra: and it was only 5 equations...
11:55 AM robotustra: with the bunch of indexii
11:57 AM robotustra: to programm some equations is one part, after you start to process data you need to make sure you are doing right thinks
11:58 AM robotustra: things*
11:58 AM edmoore: i tried to write an automatic differentiation package and that really did require some very careful thought to get it all correct and everything differentiated at the right time and not too many or too few times
11:58 AM robotustra: that's why I go simple, I take 2 frames from 2 eyes and try to rebuild 3d space
11:59 AM robotustra: edmoore, was it your video about castle visualisation?
12:00 PM edmoore: i used to work with neill campbell too, who did reconstruction from different views http://cs.bath.ac.uk/~nc537/research_my_research.html
12:00 PM edmoore: no that was qi pan
12:00 PM edmoore: my project was 3d reconstruction from a camera mounted to a robotic airship
12:00 PM robotustra: while flying?
12:01 PM edmoore: it was a specific project for the EU to monitor avalanches in the alps
12:01 PM edmoore: yes
12:01 PM robotustra: nice and cool
12:01 PM edmoore: it would fly over the same patch of avalanche field each day and estimate the volume of avalanched snow
12:01 PM edmoore: feature extraction from smooth bright white snow field is... challenging
12:01 PM robotustra: I saw some videos how quads can rebuild 3d map realtime
12:01 PM edmoore: i almost implemented a sort of bomb bay to drop a thousand bright orange ping-pong balls onto the ground
12:02 PM edmoore: but that was environmentally questionable
12:02 PM veverak: :D
12:02 PM robotustra: to map the snow?
12:02 PM edmoore: yes
12:02 PM robotustra: haha
12:02 PM edmoore: this was 2009, using a pc104 core2duo i7 cpu
12:02 PM robotustra: bb in 15 min
12:02 PM edmoore: rolls royce back then
12:08 PM edmoore: (I think i was talking poo just then, it was a core2duo, i7 came later)
12:09 PM robotustra: now I understand why you was talking about lighting and etc
12:10 PM robotustra: it's because you was wirking with grades of white :)
12:12 PM robotustra: I think for usual indoor light and with white autobalance camera it could be simplier
12:13 PM edmoore: i'm still not sure about that
12:13 PM edmoore: but i leave you to play with your technique and see what happens
12:13 PM robotustra: it's supposition
12:13 PM robotustra: yes, try and see what will happen
12:27 PM robotustra: programming cv is cool, but I started heavy hardware project recently, and I think I'll come back to CV only in 3-6 monthes form now :(
12:28 PM robotustra: I'm planning to scrape tonns of metal during long winter evenings
12:30 PM robotustra: edmoore, what country are you now?
12:34 PM edmoore: uk
12:39 PM robotustra: you was working in europe?
12:40 PM robotustra: anyway, don't answer
12:40 PM robotustra: my stupid curiosity
12:41 PM polprog: a/ca
12:41 PM polprog: oops, wrong window
12:43 PM robotustra: wrong window incontinence
12:43 PM edmoore: it was a european project with about 4 univeristies
12:44 PM edmoore: part of a hydrology monitoring thing
12:44 PM robotustra: edmoore, what area are you engineering?
12:44 PM robotustra: I downshifter to game testing robots :)
12:44 PM robotustra: downshifted*
12:45 PM edmoore: my day job is rocket propulsion research
12:45 PM edmoore: but i keep an interest in robotics, machine learning, statistic techniques for various things like signal processing and control theory, etc
12:48 PM robotustra: edmoore, are you building rocket engines?
12:49 PM robotustra: I get into R&D group where I am the only hardware guy :)
12:49 PM robotustra: and all hardware I have to build myself
12:50 PM robotustra: and program it after
12:51 PM edmoore: yes i am
12:51 PM robotustra: no pressure, no rocket science, but the set of interests is about the same, robotics, AI, cnc, electronics, etc
12:51 PM edmoore: we build and test
12:51 PM robotustra: fot ESA?
12:52 PM robotustra: for*
12:52 PM edmoore: among others certainly
12:52 PM edmoore: we have done a number of r+d projects for them
12:52 PM robotustra: spacex?
12:52 PM robotustra: nasa?
12:52 PM edmoore: no, it's very unusual for american rocket companies to work with non-americans
12:52 PM edmoore: because of ITAR
12:52 PM robotustra: ah ok
12:53 PM robotustra: you might be involved in satellites building too?
12:53 PM edmoore: a lot of my work is in air-breathing propulsion, for hypersonic flight and enabling single-stage-to-orbit
12:53 PM edmoore: not so much satellite stuff, although we have made bits-and-bobs to support sat programmes
12:53 PM edmoore: testing, fuel loading at the launch site, etc
12:54 PM edmoore: because we can do both plumbing and electronics, we get asked to make a lot of fairly niche machines in that area
12:54 PM edmoore: (e.g. machines to accurately fills sats with fuel in the clean room before they get intergrated to the launcher upper stage)
12:54 PM edmoore: it can help pay the bills between research programmes
12:55 PM robotustra: single stage - it's a bit difficult I think
12:55 PM robotustra: as soon as I know now 2 stages are usually used
12:56 PM edmoore: it's very difficult
12:56 PM robotustra: why to bring 1st stage to the orbit?
12:56 PM robotustra: it's a lot of extra fuel
12:56 PM edmoore: the idea of air-breathing engines is to reduce the amount of oxidiser you need to carry onboard
12:56 PM edmoore: which helps tremendously with mass-fraction
12:56 PM robotustra: start from aircraft?
12:57 PM edmoore: and it's the mass-fraction requirements that make it so hard to do single-stage in the first place
12:57 PM edmoore: no
12:57 PM robotustra: ah got you
12:57 PM robotustra: feed air as oxidizer
12:57 PM edmoore: just one single unit that takes off from a runway, flies to the top of the atmosphere and reaches hypersonic speeds, then runs the engines as pure rockets to leave the atmosphere and go to orbit
12:57 PM edmoore: yep
12:58 PM robotustra: nice idea
12:58 PM robotustra: kind of air turbine + rocket engine
12:58 PM edmoore: yes exactly
12:58 PM robotustra: air turbine will work till 10 km?
12:59 PM edmoore: yep, 25km is the design plan
01:03 PM robotustra: after that almost no air
01:04 PM robotustra: edmoore, are you building robots at home?
01:04 PM robotustra: because we can do both plumbing and electronics, we get asked to make a lot of fairly niche machines in that area
01:04 PM robotustra: 18:54 < edmoore> (e.g. machines to accurately fills sats with fuel in the clean room before they get intergrated to the launcher
01:04 PM robotustra: upper stage)
01:04 PM robotustra: ups
01:04 PM robotustra: sorry
01:05 PM AchiestDragon: unless you got enough power to create a "magnetic containment" field on the intake
01:05 PM edmoore: i don't do that much mechanical stuff at home, i have not the biggest house, but i have a nice workshop at work when i can use freely
01:05 PM edmoore: i do do electronics at home though
01:06 PM robotustra: edmoore, I live in app, but I'm building small cnc machines to do some stuff
01:06 PM edmoore: i have thought about it too
01:06 PM robotustra: may be with the time I get a house
01:06 PM edmoore: a cnc taig or something
01:06 PM edmoore: but we have a cnc bridgeport at work
01:07 PM edmoore: so i tend to just use that
01:07 PM robotustra: to have at work it's cool, but software companies usually don't have a machineshops
01:07 PM robotustra: what is taig?
01:09 PM robotustra: bridgeport == router?
01:09 PM robotustra: or heavy milling stuff?
01:09 PM edmoore: nope, just a normal big mill
01:09 PM edmoore: yeah
01:09 PM robotustra: ok
01:09 PM edmoore: the latter
01:10 PM edmoore: it's not heavy-milling by the standards of a big industrial vertical machining centre
01:10 PM edmoore: but for our 'toolroom' purposes it's fine
01:10 PM robotustra: i did a router, now started to make cnc lathe
01:10 PM edmoore: and can handle tricky alloys
01:10 PM edmoore: i have just been asked if i want to go for a pint
01:10 PM edmoore: it's 7pm here
01:10 PM edmoore: with colleagues
01:10 PM edmoore: i will, so i will catch you later
01:10 PM edmoore: we can continue chatting
01:11 PM robotustra: sure
01:11 PM robotustra: se you
01:11 PM robotustra: see*
01:14 PM robotustra: edmoore, btw nice to meet you
01:14 PM AchiestDragon: have fun
01:15 PM robotustra: AchiestDragon, now come back to your hexapod. do you have closeups?
01:16 PM AchiestDragon: no
01:16 PM AchiestDragon: was some time back built it in 2007
01:18 PM AchiestDragon: hopeing to get a 3d printer up and running this comming year and build another , but not decided if to go for another hexipod or a humaniod style walking robot
01:19 PM robotustra: well, I regret that I don't document some of my projects as good as I want
01:20 PM robotustra: now I'll make more pics and write the text what I did
01:20 PM robotustra: I'm building wheeled robot
01:20 PM robotustra: humanoid size
01:22 PM robotustra: something like this http://i.imgur.com/xZEKXIT.png
01:22 PM edmoore: And you robotustra
01:22 PM robotustra: but I should review the design and make it from metal
01:22 PM robotustra: at least durable materials
01:23 PM robotustra: edmoore, sure, friends are waiting for you in the pub
01:23 PM AchiestDragon: i quite like this one https://www.youtube.com/watch?v=0jyBiECoS3Q
01:23 PM edmoore: I made a hexapod for a school project, many years ago now when i was 17 and my school just got a cnc router
01:23 PM edmoore: https://www.dropbox.com/s/pc7d6w7atwmazbn/hexapod.jpg?dl=0
01:24 PM edmoore: I loved that router
01:24 PM edmoore: I am actually waiting for them! They are locking the site
01:24 PM edmoore: It’s a bit involved at a rocket test site
01:24 PM robotustra: cool
01:25 PM robotustra: this is a Al milled hexapod
01:25 PM edmoore: Yep
01:25 PM Ultrasauce: https://www.youtube.com/watch?v=2vBi6LNklRg here's mine
01:26 PM AchiestDragon: i dont realy want to use servos though , there noisy for one
01:26 PM robotustra: ah peter I spotted you
01:26 PM robotustra: canadian floors
01:27 PM edmoore: Hex’s seem like a rite of passage
01:27 PM Ultrasauce: fully static walking makes it a pretty nice entry project
01:28 PM robotustra: AchiestDragon, you told it consumes a lot?
01:29 PM robotustra: cool control
01:30 PM robotustra: AchiestDragon, what is it joystick from?
01:30 PM edmoore: Yes i tried to put foot sensors on mine for terrain adaptive whatever but was quickly out of my league aged 17
01:30 PM AchiestDragon: yea it did , 18 * 36V actuators that could shift about a metric ton each
01:31 PM robotustra: edmoore, yeah, after some ages you have to do everything by yourself and for your own money
01:31 PM edmoore: Well i just had no idea what i was doing then
01:32 PM edmoore: Had bot even learned calculus
01:32 PM robotustra: that's why I'm building home workshop
01:33 PM robotustra: edmoore, now you have ALL necessary knowledge to have a good hobby :)
01:33 PM logstaren: well not by yourself, you have us :3
01:33 PM robotustra: logstaren, who the hell you are? :)
01:34 PM logstaren: anonnumberanon
01:34 PM AchiestDragon: been looking at Flexinol Actuator Wire , seems light weight , without noise but the wire heats so going to need some thought on mountings on 3d printed plastic frame
01:34 PM Ultrasauce: also slow af
01:35 PM robotustra: AchiestDragon, I played with some NiTi threads, 1) slow 2) small contraction 3) hard to control current
01:35 PM Ultrasauce: oh that reminds me, rue_mohr did you get anywhere with that air muscle project in my absence?
01:36 PM robotustra: logstaren, I don't remember you, but it's my fault I think
01:36 PM robotustra: logstaren, do you work with edmoore ?
01:37 PM logstaren: robotustra, tried to make a baby-sized biped robot out of servos
01:37 PM logstaren: no i don't i don't know him
01:39 PM logstaren: robotustra, this is me https://www.youtube.com/watch?v=JQdXDg9PtzY&t=319s
01:39 PM AchiestDragon: my usual failing is that i desing and work using a wired psu , then seem to think about onboard power as an after thought
01:40 PM robotustra: logstaren, ah, I remember you
01:40 PM logstaren: :3
01:41 PM robotustra: AchiestDragon, what battery do you use?
01:42 PM robotustra: logstaren, how many DOF you baby robot has?
01:43 PM logstaren: it's not working right now so 0
01:43 PM logstaren: heh
01:44 PM AchiestDragon: the big one was before i had lipos and was powered by 8 led acid jell 12v 7ah yasua batteries
01:44 PM robotustra: nevertheless you did a hardware, it's anout half of the robot, now you can work on software
01:44 PM AchiestDragon: the whole thing weighed 120kg
01:44 PM robotustra: AchiestDragon, :)
01:45 PM logstaren: but it can move its arms with 3 dof per arm, 4 dof per leg, 1 for the head, so 15 dof overal
01:45 PM logstaren: yeah i was doing this experiment of testing my servo controller on it but it got burned in the process
01:45 PM logstaren: shorted, i think
01:45 PM AchiestDragon: and in the pose as on the pic it needed a 10foot dia space to stand in
01:46 PM robotustra: AchiestDragon, how did you control it?
01:46 PM logstaren: so i need to remake that board, and upgrade the software and even maybe the microcontroller because it was kind of slow for running all these 17 servos
01:47 PM robotustra: I'm making a puppet input device
01:47 PM robotustra: you can adjust joints and read values
01:47 PM robotustra: 28 DOF
01:47 PM AchiestDragon: micor atx mounted in the front with wifi acess point onboard liked to pc or laptop to work as a remote
01:47 PM AchiestDragon: micor /micro
01:48 PM logstaren: also i need some kind of massive software solution to tell how horrible the servos will react, ahead of time, since I'm dealing with horrible innacurate servos, there is flex in the hole robot, the servos are weak and innacurate, so maybe a software solution can take all of that mediocrity into consideration and actually run it all correctly
01:48 PM robotustra: monster hexapods envade canada
01:49 PM robotustra: I would not recommend to fix hardware faults with software :)
01:49 PM logstaren: heh
01:49 PM logstaren: it's just a dream
01:49 PM robotustra: my experience show me that's not right way
01:49 PM robotustra: I'd rather fix hardware
01:53 PM robotustra: you can put external encoders
01:53 PM robotustra: for instance
01:57 PM logstaren: i am weighing in two possible choices and how much work and wasted time they have, 1: use many MPU units to keep all servos aligned correctly at all times, a lot of software, and then get into the danger zone of actually trying to make the robot walk
01:57 PM logstaren: 2: user better servos, very strong and digital = a LOT of money
01:58 PM logstaren: 3: make my own digital serial servos = a LOOOOT of time
01:58 PM AchiestDragon: on the seond hexipod i was doing hardware pll loops in a fpga , ie take the positional feedback and drive the actuators to keep them in position and just feed it with the position you wanted it to move to , i had another vhdl code module to sincronise the rate of change between multiple servos , but ran out of gate space trying to do a cpu in there also
01:59 PM logstaren: options 2 and 3 will drastically reduce software needs as for needing to read MPU all the time to see if the robot is standing straight, and overcompensating for the looseness
01:59 PM AchiestDragon: seond / second .. ha too many typos
02:01 PM robotustra: AchiestDragon, which FPGA do you use?
02:02 PM robotustra: I didn't touch FPGA yet
02:02 PM robotustra: don't know if I should start to learn FPGA
02:02 PM logstaren: writing things in vhdl for a robot controller?
02:02 PM AchiestDragon: xlink 1600 gate board by enderpoint , could of done with one at least 4 times the size ,, will try and find link
02:02 PM logstaren: you gain speed and lose code writing time no?
02:03 PM logstaren: i took a class in vhdl in college programs kind of tool a long time to write iirc
02:04 PM AchiestDragon: https://www.enterpoint.co.uk/products/pga-modules/darnaw-1/
02:04 PM robotustra: logstaren, actually, you can go with other way, partially buy better servos for loaded joints and make own servos for other joints
02:05 PM AchiestDragon: one of those with a arm 32 bit cpu would of done also , but my cpu core took up about 2400 gates
02:05 PM robotustra: AchiestDragon, cool stuff, I didn't touch FPGA yet, becaus I don't know if there free software to program it
02:06 PM AchiestDragon: xilinx ise s/w used to be free , the win version was fine but the linux version sucked
02:07 PM robotustra: what about NN and self learning to walk?
02:07 PM robotustra: AchiestDragon, it's a serious con
02:07 PM logstaren: robotustra, i want to get back into it and maybe start playing more with gears, servo cases, motors, ARM, machine learning and neural networks
02:07 PM logstaren: but i can't right now
02:08 PM robotustra: logstaren, to make servos at home you need at least a router
02:08 PM robotustra: with at least 0.05mm precision
02:09 PM robotustra: AchiestDragon, how did you start with fpga stuff?
02:10 PM AchiestDragon: well the chips look good on paper , but almost everytime you use one you find you run out of space , the vertex pro for example on version comes with 8* gigbit serial ports and onboard ppc , but if you load the hdl for 8 serial connections alone you need almost double the number of avilable space than the chip has
02:10 PM logstaren: something like a small version of this? http://www.vandenbusschenv.be/files/productsmall/31/12/5axis_w.jpg
02:12 PM logstaren: 0.05mm, that's even possible?
02:12 PM AchiestDragon: from proms and eproms , to pals &gals , to cplds and fpgas , fpgas where originaly called logic cell arrays before that part of amd semiconductors became xilinx
02:13 PM AchiestDragon: done logic design as a hobby and at some points porfesionaly since bitslice
02:14 PM AchiestDragon: my typos suck , sorry athritis and rsi
02:15 PM robotustra: logstaren, yes it'a possible, I do 0.02mm at home
02:15 PM robotustra: 20 mictons
02:15 PM robotustra: for milled parts
02:15 PM AchiestDragon: its why i need to get a 3d printer as cant opperate machinary now without getting cramps
02:16 PM robotustra: why?
02:17 PM AchiestDragon: arthritus in fingers , i get cramps after a few mins when using tools
02:17 PM robotustra: logstaren, https://www.ebay.com/itm/1-5kw-VFD-cnc-6040-4-axis-router-DRILLING-MILLING-ENGRAVER-mill-MACHINE-Desktop/262324856865?_trkparms=aid%3D555019%26algo%3DPL.BANDIT%26ao%3D1%26asc%3D41375%26meid%3De262c7ab68924466b15216201d279107%26pid%3D100506%26rk%3D1%26rkt%3D1%26&_trksid=p2045573.c100506.m3226
02:18 PM robotustra: this guy can work OK
02:18 PM robotustra: but you can make your own about 20% chiper
02:19 PM robotustra: if you buy chineese one, most probably you'll need to upgrade it
02:19 PM robotustra: to get better results
02:20 PM robotustra: but for milling small stuff it's very useful.
02:21 PM logstaren: is this what you use?
02:21 PM logstaren: this is the cost of about 10 high quality strong servos by the way hehe
02:21 PM robotustra: no, I built my own
02:22 PM logstaren: you can get 0.02 with a home-built? I am impressed
02:22 PM robotustra: servos are not a tool
02:22 PM logstaren: cost you about 1000 dollars in parts?
02:22 PM logstaren: robotustra> servos are not a tool
02:22 PM logstaren: yeah i know what you mean there...
02:23 PM robotustra: https://www.youtube.com/watch?v=BwhQ6ab06e0
02:23 PM robotustra: you can build many many servos while you have a machine and did 1 design
02:24 PM robotustra: and you can improve or modify your design after
02:25 PM robotustra: logstaren, yeah, about 1200 CAD
02:25 PM robotustra: except that it's made of solid 3/4" Al plates
02:26 PM robotustra: for table at least
02:26 PM logstaren: how long to build something like this, also all i have is auto mechanic tools
02:27 PM AchiestDragon: most of the cnc "engravers" type machines are that if you want to mill or drill with them they just cant take the force and skip steps
02:27 PM logstaren: a few drills, an angle grinder, a dremel
02:27 PM robotustra: https://ic.pics.livejournal.com/maholet/24765393/9471/9471_600.jpg
02:27 PM robotustra: I had 1) files, 2) taps, 3) drill press 4) vise
02:28 PM AchiestDragon: or only work with soft materials like balsa wood effectivly
02:28 PM robotustra: calipers
02:28 PM robotustra: hacksaw
02:28 PM logstaren: ah yeah, a drill press..
02:28 PM robotustra: I actually milled a gun metal on it, but I don't want to abuse it
02:29 PM robotustra: which is not a soft material
02:29 PM robotustra: but I did it very very slowly
02:32 PM robotustra: AchiestDragon, If you have a direct drive - they probably will loose steps, I did reduction 1:3 and I use 1/2-10 ACME screw
02:32 PM robotustra: so I have 6000 steps per inch
02:32 PM robotustra: it's without mictostepping
02:33 PM robotustra: which is 4 microns
02:33 PM logstaren: if you have a guide to make one like yours don't hesitate to post it here!
02:34 PM robotustra: let's say I 5 time less precise because of rigidity
02:35 PM robotustra: logstaren, as I told, I poorly documented it and I regret about it
02:35 PM robotustra: but I can explain you how to make even better than mine
02:35 PM robotustra: for the same money
02:36 PM robotustra: actually Tom_L is building one for himself
02:36 PM logstaren: ah that's what he was doing the other day
02:37 PM robotustra: after I built my cncn router I got how ugly were my hand filing parts :)
02:37 PM AchiestDragon: i have a amadeal xj25 mill , but its not suitable for cnc conversion without a lot of improvments , a 3 d printer will be more usefull
02:38 PM robotustra: AchiestDragon, after hexapod it will be easy to build 3d printer
02:38 PM robotustra: AchiestDragon, what kind of printer do you want?
02:39 PM robotustra: AchiestDragon, it's quite big mill
02:39 PM orlock: use the printer to make mounts for the steppers to CNC the mill
02:39 PM robotustra: :)
02:39 PM orlock: AchiestDragon: Nice, wish i had something like that
02:39 PM AchiestDragon: well i was looking at the tevo delta monster as 1 it has good reviews and 2 since its my son that realy wants me to get back into robots seems a good idea so i can get him to assemble it and what i make on it
02:40 PM robotustra: or use your mill to mill a 3d printer parts
02:40 PM orlock: Hah, interesting they sell stuff branded as "Syil"
02:40 PM orlock: here, and from the factory, its got a different brand
02:40 PM orlock: "Seig"
02:42 PM robotustra: I approve deltas
02:42 PM robotustra: like delta printers
02:42 PM AchiestDragon: the mill cant be used where it is and i cant move it back in the flat here as its too noisy to use in here , but i could get away with a 3d printer
02:43 PM orlock: yeah, as long as you are ok with the smells
02:43 PM orlock: PLA is ok
02:43 PM orlock: ABS is like melting lego
02:44 PM AchiestDragon: the tevo would look kool here
02:45 PM AchiestDragon: yea PLA for that reason also , the one i built was for abs but its vented where it is , just that the 6 foot square outhouse is in is too full with other crap to be able to work on it
02:46 PM robotustra: AchiestDragon, that's why I keep my mill in the sound proof box
02:46 PM robotustra: ABS is warping a lot
02:46 PM AchiestDragon: and it also looks like it was thown together with a buldozer
02:47 PM robotustra: I like PLA more
02:48 PM robotustra: I don't know, my ABS prints were OK, i just don't like that it shrinks a lot and it's cracky
02:50 PM robotustra: PLA is much more risistant for mechanical loads
02:50 PM AchiestDragon: was looking did you know theres a 3d printer that boasts its good enough to be able to print industreal strength nylon, but it was over £6k
02:51 PM orlock: AchiestDragon: dunno what that means any why a normal printer would not do it
02:51 PM orlock: $5 hot glue guns extrude nylon
02:51 PM AchiestDragon: i have even thought about just sticking the mig welder nozzle to the end of it and look at using it that way , shame i bothered to use a 5mmthick stainless steel sheet for the heated bed
02:52 PM AchiestDragon: about 350C or something like
02:52 PM AchiestDragon: cant remeber but a lot more than abs or pla
02:53 PM orlock: also, lots of companies lie about having something special
02:53 PM orlock: and will bump up the price stupidly
02:53 PM orlock: :-(
02:54 PM orlock: makes me sad
02:54 PM AchiestDragon: yea
02:54 PM robotustra: actually, you can print nylon on any printer, you just need to upgrade your hot end to fullmetal
02:55 PM robotustra: and be able to handle 110C on your heat bed
02:55 PM robotustra: not 350, only 280C
02:56 PM robotustra: but there are better plastics now like polycarbonate
02:56 PM robotustra: they are even stronger than PLA
02:57 PM AchiestDragon: it took 45 mins for the bead to heat to 110c thats a 650mm * 360mm 5mm thick sheet of sainless steel using 48v at 40A ,
02:58 PM robotustra: ha ha
02:58 PM robotustra: monster printers
02:59 PM AchiestDragon: i was using a plugable octal relay , it melted that and the socket , i have a 3phase contactor to replace that with but i also blew up the psu
02:59 PM robotustra: make you own heater from tungsten wire
03:00 PM robotustra: heh
03:00 PM robotustra: don't burn your wooden house :)
03:01 PM AchiestDragon: yea neo wire taped to the underside , with the right tape
03:02 PM AchiestDragon: actualy supprized that is still stuck its been in a damp environment for the last 4 years and no signs of it dropping off
03:03 PM robotustra: or buy this one https://www.ebay.com/itm/Aroma-Kitchen-Single-Hot-Plate-Dinner-Portable-Electric-Countertop-Burner/182478707923?hash=item2a7c941cd3:g:e6EAAOSwax5YvPCz + PWM
03:14 PM AchiestDragon: better than the istove then http://www.trbimg.com/img-532ed009/turbine/la-fi-tn-istove-apple-knockoff-china-20120227
03:16 PM AchiestDragon: bet it makes a better coffee
03:18 PM robotustra: :)
03:23 PM robotustra: better or bitter
03:38 PM AchiestDragon: well if you use one as a phone a bet your ears would be burning
05:05 PM logstaren: oh god this is pr0n https://www.youtube.com/watch?v=8NatiYBitAI
05:56 PM robotustra: logstaren, I see you already started to think right direction
05:58 PM robotustra: ta-da https://imgur.com/a8HfDPP
05:58 PM robotustra: 4 hours, 4x4"
06:00 PM robotustra: fortunately, I don't have lapping plate at home
06:01 PM logstaren: robotustra, a somewhat flat, apparently steel plate, with a hole in it in a random position, what' s the story?
06:01 PM robotustra: I scraped it in 4 hours to this state
06:02 PM robotustra: this is a beginning of my cnc lathe :)
06:02 PM logstaren: ah ok
06:02 PM logstaren: yeah lathe
06:02 PM robotustra: have more metal to scrape
06:02 PM robotustra: https://imgur.com/e9M8PyS
06:03 PM logstaren: oh i have a good welder now :)
06:03 PM robotustra: all adjessent surfaces must be flat, parallel or square
06:03 PM robotustra: I don't have that's why I'll bolt it
06:04 PM robotustra: I know how weld could stratch metal
06:04 PM robotustra: it's neally impossible to weld precisely
06:05 PM robotustra: +- 0.5 mm
06:05 PM logstaren: for what job? for sticking 2 pieces of metal together permanently it is great
06:06 PM logstaren: tack weld the project first and measure your clearances, if happy, apply stronger welds
06:06 PM logstaren: if not happy, grind the small tack welds away and reposition the pieces
06:07 PM robotustra: my tolerances for the frame fill be 10 microns for parellelness
06:07 PM robotustra: for 700 mm
06:08 PM robotustra: and <10 microns for 200 mm for squareness
06:08 PM robotustra: not sure I can do it with welder at home
06:09 PM robotustra: bolts are good because I can adjust it easily
06:10 PM logstaren: right
06:10 PM logstaren: im looking at making a cnc table but i don't want it to become a full time hobby and get in the way of my robot project
06:11 PM logstaren: i don't know how to look at it this way
06:11 PM robotustra: I did all metalwork in 16 weekends
06:12 PM robotustra: do you know what is BNF?
06:12 PM logstaren: no
06:12 PM robotustra: https://en.wikipedia.org/wiki/Backus%E2%80%93Naur_form
06:13 PM robotustra: consider tool making as a necessary process
06:14 PM robotustra: actually the only thin humans doing could be written in very simple formula in BN form: {make a tool to}+
06:15 PM robotustra: i.e. make a tool to make a tool to make a tool ....
06:16 PM robotustra: so, if you are not making a tool - what are you donig then?
06:17 PM logstaren: I don't have a come back to that.
06:17 PM robotustra: come back to what? :)
06:18 PM logstaren: I mean I don't have a good answer to that question.
06:18 PM robotustra: :)
06:40 PM * Tom_L thinks robotustra is a tool
06:41 PM rue_mohr: really, you guys been on about 3d printers the whole day?
06:42 PM Tom_L: hey
06:42 PM Tom_L: don't knock it
06:42 PM Tom_L: at least the channel wasn't idle all day
06:42 PM logstaren: im sorry did i do something wrun?
06:45 PM Tom_L: no
06:45 PM Tom_L: or you wouldn't be here
06:45 PM Tom_L: :)
06:45 PM logstaren: machines are boring
06:45 PM logstaren: :3
06:51 PM logstaren: rue_mohr, well since nobody ever offered to make me servo cases i had to look up how to make them
06:51 PM rue_mohr: I didn't se you ask
07:07 PM rue_mohr: for years I been using the 2N2907 as the compliment for a pn2222
07:07 PM rue_mohr: wow was I shocked when I looked up its specs
07:08 PM rue_mohr: so, I'm gonna migrate to 2n4401 / 2n4403
08:35 PM logstaren: man... this dude... what is this i need a japanese friend https://www.youtube.com/watch?v=xUDIYxYhd2o
08:43 PM logstaren: man here is a translation of his article why didn't I do this sooner?? https://translate.google.com/translate?hl=en&sl=auto&tl=en&u=http%3A%2F%2Fai2001.ifdef.jp%2Fai%2Fai.html
08:47 PM robotustra: yeah his robot walks perfectly
08:47 PM logstaren: perfect is the word
08:48 PM logstaren: lol that translation if funky!
08:48 PM robotustra: as soon as I understand, he trained the brain of his robot with the model of inversed pendulum
08:49 PM robotustra: you can do it also
08:51 PM logstaren: ok lemme finish the article then i wanna talk about this
08:52 PM logstaren: all i know for doing balancing is for a quadcopter with an MPU chip
09:09 PM logstaren: interesting read
09:20 PM Luminax-Work is now known as LuminaxWk