#robotics Logs

Dec 08 2018

#robotics Calendar

05:39 AM veverak: so
05:43 AM veverak: I have N-dimensional vector space
05:43 AM veverak: I have projection of that space into 3D space
05:43 AM veverak: in the 'N'-space I have a set of points forming a convex hull
05:43 AM veverak: I need to increase the number of points in the shape in 'N'-space, in a way, that there is at leas tcertain density of points after the projection
05:43 AM veverak: hmm
05:43 AM veverak: If I split the 'N'-dimensional convex hull into simplexes, that would make stuff easier
05:48 AM veverak: seems splitting hull into simplexes is not easy either
05:58 AM deshipu: vetices
05:58 AM deshipu: vertices
05:58 AM deshipu: not vertexes
06:01 AM veverak: ah :)
06:01 AM veverak: wait
06:01 AM veverak: it was supposed to be 'voxels'
06:01 AM veverak: fu me
06:02 AM veverak: so, what I actually need is to project N-dimensional set of points that forms convex hull into 3D space and mark all voxels it collides with
06:03 AM veverak: hmm, not just collides, completely overlaps
06:05 AM veverak: (that N-dimensional space is configuration space of robotic arm - vector of angles, that projection is 'kinematics' - position of the end of effector of that arm)
07:06 AM rue_bed: didn't you already do that with the tree structure?
07:06 AM veverak: this one nope
07:06 AM rue_bed: the one with 14 leaf nodes
07:06 AM veverak: and the tree does not help here
07:07 AM rue_bed: huh
07:07 AM veverak: 14???
07:07 AM rue_bed: each level of the tree breaks a cubeic space into 1/n spaces
07:07 AM rue_bed: the more levels you add, the more it defines the space
07:07 AM rue_bed: er shape
07:08 AM rue_bed: so, who was it that did that?
07:08 AM rue_bed: 27leaves I suppose makes sense
07:08 AM rue_bed: its a wonderfull idea, its not mine
07:09 AM veverak: it is
07:09 AM veverak: I use that
07:09 AM veverak: is not relevant to this problem really
07:09 AM veverak: I mean, I use it, but it's not the problem :)
07:09 AM rue_bed: some already did it
07:09 AM veverak: but I want to create it, yeah
07:09 AM veverak: idea is to have tree that represents the space the robot arm came move in
07:16 AM veverak: -> each leaf voxel is a space where arm can move
09:08 AM deshipu: voxel is a 3d pixel
09:26 AM veverak: we know :)
09:30 AM deshipu: I'm just making sure I understand what you are saying
09:30 AM deshipu: so you are basically discretizing the space and trying to solve the problem in that?
09:31 AM deshipu: optimization problems are easier to solve in non-discrete spaces
09:31 AM deshipu: especially with convex areas
10:01 AM veverak: it's not an optimization
10:01 AM veverak: also
10:01 AM veverak: "optimization problems are easier to solve in non-discrete spaces" is quite a wild statement :)
10:04 AM veverak: deshipu: what I need, so easily detect if something is in range of working space of that arm
10:04 AM veverak: => it's in collision with the area of all possible positions of end effector
10:06 AM veverak: 'octree' (tree of voxels) is quite an effective representation of an area in terms 'detection of collision'
11:19 AM veverak: anyway, I have solution
11:19 AM veverak: I can overshoot the resolution of the thing that generates the datapoints easily
11:19 AM veverak: so it should be doable to have distance of points after projection smaller than is voxel size
11:20 AM veverak: I can also mark 'corner' points of the hull - also mark all voxels in collision with that points
11:20 AM veverak: marked voxels will be removed after the entire process
11:20 AM veverak: => win \o/
12:39 PM rue_bed: well who already did it
12:40 PM rue_bed: Tom_L, would you mind searching the logs for 'voxel'?
12:40 PM rue_bed: the result is from a year or two ago
12:40 PM mrdata__: that's way past my scrollback horizon
12:41 PM rue_bed: we have a log tho
12:41 PM rue_bed: tom can search them
12:41 PM rue_bed: if we knew who did it, maybe they will share their code
12:42 PM rue_bed: Tom_L, how big are all the logs?
12:42 PM rue_bed: maybe I should buy a thumb drive and get you to send me a copy
02:47 PM justanotheruser is now known as Guest39745
02:51 PM justan0theruser is now known as justanotheruser
02:53 PM veverak: deshipu: https://pixycam.com/pixy2/ if you are interested...
03:29 PM deshipu: veverak: that's even more overpriced than openmv
03:30 PM veverak: seen it in RL, it works perfectly
03:30 PM veverak: (tje line following)
03:30 PM veverak: will stresstest it more :)
03:33 PM deshipu: veverak: I'm generally sceptical about using atmegas in modern projects
03:33 PM veverak: true to that
03:34 PM deshipu: skeptical
03:34 PM deshipu: sry
03:34 PM veverak: Processor: NXP LPC4330, 204 MHz, dual core
03:34 PM veverak: ^^ this is in pixy
03:35 PM veverak: I think this could do pixel flow perfectly
03:36 PM deshipu: just need to program it...
03:36 PM deshipu: "just"
03:36 PM veverak: wat, why?
03:36 PM veverak: depends on what you want
03:37 PM veverak: oh, coding the pixel flow
03:37 PM veverak: yeah, but I thought that learning the algo behind pixel flow could be interesting
03:37 PM veverak: (either for me or as bc thesis for osme student)
03:38 PM deshipu: you can look at openmv's sources to see how they did it
03:38 PM veverak: yup
03:39 PM deshipu: btw, mouse sensor with a lense would be much cooler
03:40 PM veverak: depends
03:40 PM veverak: deshipu: I am actually interested in trying different way to get sense of location for the robot
03:40 PM veverak: encoders, optical flow, imu
03:40 PM veverak: and merge them with kalman filter
03:40 PM veverak: :)
03:41 PM deshipu: ugh
03:41 PM deshipu: add gps and wifi location ;)
03:42 PM veverak: !!!
03:42 PM veverak: :)
03:42 PM veverak: the point: merging multiple sources of location in sensible way
03:42 PM deshipu: oh, and a wide-angle up-pointing hi-res camera for navigating using the stars
03:43 PM veverak: that's optical flow too no ?
03:43 PM veverak: ah, stars specifically
03:44 PM deshipu: I mean matching to constellation patterns and doing the math
03:44 PM veverak: cool
03:44 PM deshipu: they are too far away for optical flow
03:45 PM deshipu: I need to try beacons one of those days
03:46 PM veverak: IT WORKS
03:46 PM veverak: leg workarea fixed
03:46 PM veverak: deshipu: you mean wifi?
03:46 PM veverak: :)
03:46 PM deshipu: ultrasound + IR
03:47 PM deshipu: I want an attiny waiting on an ir signal, and sending a ultrasound pulse when it gets it
03:47 PM veverak: sounds cool
03:47 PM deshipu: as the beacon, with different ir codes for different beacons
03:48 PM deshipu: then the robot would have an omnidirectional ir led for sending the signals, and an ultrasound sensor for receiving the pings
03:48 PM deshipu: should work resonably well for things like museums
03:49 PM veverak: definetly
03:50 PM deshipu: the ir sensor is rather low-power
03:50 PM deshipu: so should last on battery
03:52 PM veverak: cool
03:52 PM veverak: deshipu: have I shown you my toy?
03:52 PM veverak: here it is!
03:52 PM veverak: https://www.youtube.com/watch?v=lSrKH6ttxWI
03:53 PM deshipu: yeah
03:53 PM deshipu: looks great
03:54 PM deshipu: hmm, if not johnny 5, then you could make a wall-e
03:54 PM deshipu: just put a box on it
03:54 PM veverak: yup
03:55 PM veverak: that's in TODO
03:55 PM veverak: :)
03:55 PM veverak: deshipu: I was thinking about gettimg him small fire extinguisher
04:25 PM deshipu: cute
04:53 PM krunk is now known as Guest20190
04:54 PM Guest20190 is now known as dravine
06:59 PM veverak: got it
06:59 PM veverak: deshipu: https://imgur.com/a/Oiujado
06:59 PM veverak: on the right: configuration space of one leg
06:59 PM veverak: on the left: workarea of other one
07:07 PM veverak: next on TODO list: interactive markers to set actual position of robot and goal position of robot
07:54 PM JaymesBeakerbott is now known as l
07:54 PM l is now known as lmnop
07:55 PM lmnop is now known as n
07:55 PM n is now known as z
07:55 PM z is now known as ZachMorris
08:10 PM rue_mohr: hahahaha sleection list have a bug in motif!
08:10 PM rue_mohr: hahahahahahahaha
08:11 PM rue_mohr: it prevents me from being able to programatically select more than one item if the selection mode for the list is set to a mode where I should be able to (you can with the mouse) select more than one item at a time
08:11 PM rue_mohr: DAMNIT
08:11 PM rue_mohr: if (((lw->list.SelectionPolicy == XmSINGLE_SELECT) ||
08:11 PM rue_mohr: (lw->list.SelectionPolicy == XmBROWSE_SELECT) ||
08:11 PM rue_mohr: (lw->list.SelectionPolicy == XmEXTENDED_SELECT)))
08:12 PM rue_mohr: XmEXTENDED_SELECT should NOT be in that list
09:01 PM flav0r: dudes
09:01 PM flav0r: life was so good !
09:02 PM flav0r: about 1mm to go in a 40 hour print !
09:02 PM flav0r: CLOGGED !!!!!!!!!
09:04 PM rue_mohr: burn
09:04 PM flav0r: i want to burn it all to the ground
09:04 PM rue_mohr: can you fake the last mm with the help of some glue?
09:04 PM flav0r: well not to worry
09:05 PM flav0r: well i can just adjust the size of the part that mates with it so no big deal
09:05 PM rue_mohr: you could slice-off the top and just print that
09:05 PM flav0r: a lid more or less
09:05 PM flav0r: well on the side in question its the open end of a “box” pretty much
09:05 PM rue_mohr: I think I'm gonna give up on my CAM program
09:05 PM rue_mohr: to do this I have to write my own gui
09:06 PM rue_mohr: I'm exerting the same amount of effort into getting around motif bugs and shortstops that I would writing my own stuff from scratch
09:06 PM rue_mohr: and I wont have anything to show for it
09:07 PM rue_mohr: this SUCKS
09:11 PM flav0r: i just measured
09:11 PM flav0r: 1.01 mm !
09:11 PM rue_mohr: just print that and glue iton
09:11 PM rue_mohr: bit of bondo and soem paint
09:11 PM flav0r: well as i was saying
09:12 PM flav0r: the surface gets covered by a “lid”
09:12 PM rue_mohr: look, the thing that I just gave up on took me more than a month to put togethor
09:12 PM rue_mohr: you got it easy
09:12 PM flav0r: so really i can just glue in the brass screw inserts
09:12 PM flav0r: and grind those down by 1mm
09:12 PM rue_mohr: ok, no more coding for a month, f*** it
09:12 PM rue_mohr: so some real crap
09:12 PM flav0r: the goal was to make sure the thing was viable as a whole
09:12 PM flav0r: yeah dude
09:12 PM flav0r: take a break
09:13 PM flav0r: do something else !
09:13 PM rue_mohr: its not a break, its a stop
09:13 PM rue_mohr: when I stop working on a projet liek that I NEVER go back
09:13 PM rue_mohr: you cant, if you dont ahve it all in your head you just cant
10:31 PM rue_shop5: if I can make anything, what do I want to make and why
10:31 PM rue_shop5: I need some drive here