#robotics Logs

Jan 24 2018

08:36 AM polprog is now known as gorplop
08:49 AM gorplop is now known as polprog
09:32 AM robotustra: morning
09:46 AM eraserpencil: hi
09:48 AM eraserpencil: has anyone tried using DFT for localising?
09:50 AM robotustra: DFT = density functional theory?
09:50 AM eraserpencil: discrete fourier transform
09:51 AM robotustra: localisation of what?
09:51 AM eraserpencil: amcl poses
09:51 AM robotustra: formulate your task pls
09:51 AM robotustra: without abbreviations
09:52 AM robotustra: people could not read your mind
09:54 AM eraserpencil: was following this paper(https://bib.irb.hr/datoteka/816062.warehousing_localization.pdf) on achieving high accuracy on localisation by means of adding Iterative Closest Point and Discrete Fourier Transform after achieving Adaptive Monte Carlo Locallisation
09:56 AM robotustra: ok
09:57 AM robotustra: what accuracy you want to achive?
09:57 AM robotustra: what accuracy do you want to achive?
09:57 AM eraserpencil: was tripping up towards the end of the math and couldnt replicate the results. A mysterious N that appeared made me suspect Nyquist Theorem was used too but was wondering is there are experiences with such implementations
09:58 AM eraserpencil: I could achieve accuracy of 5cm, but i really need +- 1cm
09:59 AM robotustra: what is relative error? 5% 0.5%?
10:00 AM robotustra: N is the number of distance mesurements?
10:03 AM robotustra: you mean that increasing of N does not lead to increasing of position measurement accuracy?
10:04 AM eraserpencil: no
10:04 AM eraserpencil: strangely towards the end, I'd get the results they got if N=2
10:04 AM eraserpencil: but throughout the paper N=1440 (number of measurements) make sense
10:07 AM robotustra: I didn't work close with that
10:07 AM robotustra: how do you want to use DFT to improve your measurements?
10:08 AM eraserpencil: it's a shot in a dark anyway
10:08 AM robotustra: are you solving some practical tasks like robot's navigation and docking?
10:08 AM eraserpencil: yea
10:08 AM robotustra: ok
10:09 AM robotustra: so you need +-1cm to get into the dock?
10:10 AM eraserpencil: there are other tasks that requires +- 1cm
10:10 AM robotustra: I think if you are in 2D space - the minimal number of mesurements you need to get the coordinates is 2
10:10 AM robotustra: so N=2 has a reason
10:11 AM eraserpencil: could you elaborate?
10:11 AM robotustra: no
10:12 AM robotustra: I have my own tasks to solve
10:12 AM robotustra: but article is interesting, I'll read it later
10:12 AM eraserpencil: alright!
10:12 AM eraserpencil: thanks for the help
10:12 AM eraserpencil: i'll chew on N=2 further
10:13 AM robotustra: but the question is, to make DFT they need much more than 2 measurements
10:14 AM robotustra: that's why I can't say anything before I read and understand what is in the paper
10:14 AM eraserpencil: oh i get it
10:14 AM eraserpencil: yea 3 is the minimum
10:15 AM eraserpencil: have fun
10:15 AM robotustra: thanks
10:16 AM robotustra: do you know the errors of your sensors?
10:16 AM robotustra: absolute error?
10:17 AM eraserpencil: i cant rmb
10:17 AM robotustra: it may happen that if the absolute sensor error is higher the accuracy you want to get from math
10:17 AM robotustra: in this case you can't get what you want
10:17 AM robotustra: I would start with that
10:18 AM eraserpencil: I was doing the math in simulation before integrating on the bot i.e drawing a 5x 5 square and generating 1440 lines for 2 points in the square
10:18 AM eraserpencil: thats the only way i could test the results shown on the paper
10:19 AM robotustra: so, you mean that the distance measurements have no errors?
10:19 AM eraserpencil: oh yea, i corrected the absolute error with odom feedback
10:19 AM robotustra: ok
10:19 AM robotustra: is your robot moving?
10:19 AM eraserpencil: yes
10:20 AM robotustra: what is the frequency of taking measurements?
06:32 PM rue_: 3.16e-10 Hz
07:34 PM DIRT: is this a random robotics channel?
07:52 PM rue_: random?
07:52 PM rue_: random as in unpredictable?
07:52 PM DIRT: yeah what goes on here?
07:52 PM rue_: radom like crazy?
07:53 PM rue_: lots of things go on here
07:53 PM rue_: but its like Schrodners cat
07:53 PM rue_: it only happens when the lid is closed
07:53 PM rue_: if you open the lid, you observe nothing
07:58 PM DIRT: Ever see the flying cat drone?
07:59 PM rue_: some things I wish I could unsee
07:59 PM rue_: that on my list
08:05 PM Tom_L: rue_ that's a fact
08:10 PM Tom_L: no progress tonight
08:14 PM rue_: I cut the ground steak, I tapped it into the gound enough to hold it there before hitting my finger and deciding to finish later
08:15 PM rue_: my brain needs to rest
08:15 PM rue_: (grrr)
08:16 PM rue_: I will try to focus on parts of an artificial brain that can be of use to me without the rest
08:16 PM justan0theruser is now known as justanotheruser
08:16 PM rue_: I want to start the list with the physical sections of a brain
08:17 PM rue_: ... anyone wanna let me disassmble their brain?
08:20 PM orlock: rue_: seen Chappie?
08:21 PM * Tom_L takes away rue_'s steak pounding card
08:38 PM anonnumberanon: rue_, get a rat
08:47 PM DIRT: Any of you use slam before?
09:18 PM anonnumberanon: done a lot of research about it
09:18 PM anonnumberanon: it's hard
09:18 PM anonnumberanon: and magic, when it works
09:53 PM Plazma: Dioz should never have kids
11:27 PM rue_: hmm
11:27 PM rue_: efficient effective
11:29 PM rue_: my fourier code calculates the numbers really quickly
11:29 PM rue_: but
11:29 PM rue_: it twiddles its thrumbs a LOT while waiting for the data
11:56 PM DIRT: I want to use a mon cam t do slam.
11:56 PM DIRT: there is a really good slam called orbslam2. But im not sure how that would run on a robot running a raspberry pi
11:57 PM DIRT: Throw this $3 keyboard away.....