Loading ...
Sorry, an error occurred while loading the content.
 

Re: [tekkotsu_dev] WalkCalibration

Expand Messages
  • Ethan Tira-Thompson
    ... Well, I guess that would be me. ;) ... All of the distance measurements should be in centimeters. I meant to include that in the inline help menu, oops.
    Message 1 of 9 , Mar 1, 2004
      Who have experience on walk calibration?
      Well, I guess that would be me. ;)

      What is the dimension of distance to be input while forward/strafe
      calibration in Cartesian mode? mm or cm?
      All of the distance measurements should be in centimeters.  I meant to include that in the inline help menu, oops.  That will be updated in my next checkin.
      Speeds are specified to the underlying walk as millimeters/second, but i didn't want to imply that such accuracy in measurements would be expected, or possible.  Although you can enter fractional amounts (e.g. 50.5) if you want.

      when i input either 50 or 500, the ControllerGUI hang and
      output "invaild input".
      So, this is pretty odd.  It does bring up a bug that the Controller doesn't currently have a mechanism for noticing when you close a window, so after the first warning message, it won't know that you've clicked 'ok', and so it won't display any more pop-up warning messages.  I'll have to think about that for a bit.  For other windows it's just been a minor nuisance of restarting the server on the Aibo, but for GUI error messages it's a significant problem.  Regardless, the error message is also printed on the console (through serr), so you can look for it there.

      But I can't get either the 'invalid input' from a number, nor the hang to occur.  The invalid input should be followed by actual text of what it thinks is invalid.  All you need to type is the number - The input is passed directly to the strtod() library function, and I'm just looking for an error from that call.  Even when I enter garbage, it just complains and waits for me to try again, hopefully that's all you meant.  (you can still click the "back" button at the bottom of the Controller to get out of the interactive calibration system.)


      I think this next part might give the clue we need though:
      Also, it can not Save Data Set which have the following error message

      fs.txt'...a to '/ms/test
      Could not open file for savingSoundManager::LoadBuffer() of 11202
      bytes
      fr.txt'...a to '/ms/test
      sr.txt'...open file for savingSaving data to '/ms/test
      br.txt'...open file for savingSaving data to '/ms/test
      bs.txt'...open file for savingSaving data to '/ms/test
      rr.txt'...open file for savingSaving data to '/ms/test
      Those filenames are wrapping funny - the beginning of the error message ("Could not open") is being overwritten by the end of the filename ("xx.txt'...")
      I'm betting you're entering the filename from the sout console port (10001) or more generally, through a telnet program, which uses '\r\n' line endings when you hit enter instead of just '\n'.  And since the '\r' character is not valid for filenames, the error ensues.

      In the short run, use the "send input" field of the Controller.  In the long run, I've checked in an updated Controller.cc to fix the problem in the console text processing callback.

      So that could also explain the problem you were having entering values earlier if you were also using the console for that.

      Hope that fixes things.
      -ethan

      Patches:
      Fixes console input:

      http://cvs.tekkotsu.org/cgi-bin/viewcvs.cgi/Tekkotsu/Behaviors/Controller.cc.diff?r1=1.35&r2=1.36
      Better help text:
      http://cvs.tekkotsu.org/cgi-bin/viewcvs.cgi/Tekkotsu/Behaviors/Controls/WalkCalibration.cc.diff?r1=1.7&r2=1.8

    • yuewah99
      I am using either Tekkotus and Tekkotsu CVS version. In the ControllerGUI, when it request to input the forward distance during the Forward/Strafe Interactive
      Message 2 of 9 , Mar 2, 2004
        I am using either Tekkotus and Tekkotsu CVS version.

        In the ControllerGUI, when it request to input the forward distance
        during the Forward/Strafe Interactive Calibration. I typed "50"
        in "Send Input" Textfield and followed by <Enter>.

        The warning message is pop up without any description and the
        ControllerGUI is hang.

        console output of ControllerGUI:

        <Error>
        >Invalid input:

        However, if i input "!", warning message is pop up. But, i input "50"
        again without hang.

        For the saving data set problem, i am using the "Send Input"
        Textfield in ControllerGUI for the filename instead of system console
        port 10001. It can save file if i hard code the filename in the
        source code.


        --- In tekkotsu_dev@yahoogroups.com, Ethan Tira-Thompson <ejt@a...>
        wrote:
        > > Who have experience on walk calibration?
        >
        > Well, I guess that would be me. ;)
        >
        >
        > > What is the dimension of distance to be input while
        forward/strafe
        > > calibration in Cartesian mode? mm or cm?
        >
        > All of the distance measurements should be in centimeters. I meant
        to
        > include that in the inline help menu, oops. That will be updated
        in my
        > next checkin.
        > Speeds are specified to the underlying walk as millimeters/second,
        but i
        > didn't want to imply that such accuracy in measurements would be
        > expected, or possible. Although you can enter fractional amounts
        (e.g.
        > 50.5) if you want.
        >
        >
        > > when i input either 50 or 500, the ControllerGUI hang and
        > > output "invaild input".
        >
        > So, this is pretty odd. It does bring up a bug that the Controller
        > doesn't currently have a mechanism for noticing when you close a
        window,
        > so after the first warning message, it won't know that you've
        clicked
        > 'ok', and so it won't display any more pop-up warning messages.
        I'll
        > have to think about that for a bit. For other windows it's just
        been a
        > minor nuisance of restarting the server on the Aibo, but for GUI
        error
        > messages it's a significant problem. Regardless, the error message
        is
        > also printed on the console (through serr), so you can look for it
        > there.
        >
        > But I can't get either the 'invalid input' from a number, nor the
        hang
        > to occur. The invalid input should be followed by actual text of
        what
        > it thinks is invalid. All you need to type is the number - The
        input is
        > passed directly to the strtod() library function, and I'm just
        looking
        > for an error from that call. Even when I enter garbage, it just
        > complains and waits for me to try again, hopefully that's all you
        > meant. (you can still click the "back" button at the bottom of the
        > Controller to get out of the interactive calibration system.)
        >
        > I think this next part might give the clue we need though:
        >
        > > Also, it can not Save Data Set which have the following error
        message
        > >
        > > fs.txt'...a to '/ms/test
        > > Could not open file for savingSoundManager::LoadBuffer() of 11202
        > > bytes
        > > fr.txt'...a to '/ms/test
        > > sr.txt'...open file for savingSaving data to '/ms/test
        > > br.txt'...open file for savingSaving data to '/ms/test
        > > bs.txt'...open file for savingSaving data to '/ms/test
        > > rr.txt'...open file for savingSaving data to '/ms/test
        >
        > Those filenames are wrapping funny - the beginning of the error
        message
        > ("Could not open") is being overwritten by the end of the filename
        > ("xx.txt'...")
        > I'm betting you're entering the filename from the sout console port
        > (10001) or more generally, through a telnet program, which
        uses '\r\n'
        > line endings when you hit enter instead of just '\n'. And since the
        > '\r' character is not valid for filenames, the error ensues.
        >
        > In the short run, use the "send input" field of the Controller. In
        the
        > long run, I've checked in an updated Controller.cc to fix the
        problem in
        > the console text processing callback.
        >
        > So that could also explain the problem you were having entering
        values
        > earlier if you were also using the console for that.
        >
        > Hope that fixes things.
        > -ethan
        >
        > Patches:
        > Fixes console input:
        > http://cvs.tekkotsu.org/cgi-
        bin/viewcvs.cgi/Tekkotsu/Behaviors/Controller.cc.diff?r1=1.35&r2=1.36
        > Better help text:
        > http://cvs.tekkotsu.org/cgi-
        bin/viewcvs.cgi/Tekkotsu/Behaviors/Controls/WalkCalibration.cc.diff?
        r1=1.7&r2=1.8
      • Ethan Tira-Thompson
        I m afraid I m stumped... I thought perhaps this is a windows specific thing, but it seems to be working for me, even on windows. I m 99% sure the problem is
        Message 3 of 9 , Mar 2, 2004
          I'm afraid I'm stumped... I thought perhaps this is a windows specific thing, but it seems to be working for me, even on windows.  I'm 99% sure the problem is that you're getting a '\r' character at the end of your inputs, but I have no idea how it's getting there from the GUI, or why it's not being stripped even if it was there.

          here's a little tip though - the input processing code can handle two inputs - so instead of entering the two measurements separately, just put some whitespace between them and it'll use the second word as the second input.  I'm betting this will let you get around having the error pop up.

          if you figure out what's causing this (and better yet, know how to fix it), let me know.

          sorry,
          -ethan

          On Tue, 2004-03-02 at 04:39, yuewah99 wrote:
          I am using either Tekkotus and Tekkotsu CVS version.

          In the ControllerGUI, when it request to input the forward distance
          during the Forward/Strafe Interactive Calibration. I typed "50"
          in "Send Input" Textfield and followed by <Enter>.

          The warning message is pop up without any description and the
          ControllerGUI is hang.

          console output of ControllerGUI:

          <Error>
          >Invalid input:

          However, if i input "!", warning message is pop up. But, i input "50"
          again without hang.

          For the saving data set problem, i am using the "Send Input"
          Textfield in ControllerGUI for the filename instead of system console
          port 10001. It can save file if i hard code the filename in the
          source code.


          --- In tekkotsu_dev@yahoogroups.com, Ethan Tira-Thompson <ejt@a...>
          wrote:
          > > Who have experience on walk calibration?
          >
          > Well, I guess that would be me. ;)
          >
          >
          > > What is the dimension of distance to be input while
          forward/strafe
          > > calibration in Cartesian mode? mm or cm?
          >
          > All of the distance measurements should be in centimeters.  I meant
          to
          > include that in the inline help menu, oops.  That will be updated
          in my
          > next checkin.
          > Speeds are specified to the underlying walk as millimeters/second,
          but i
          > didn't want to imply that such accuracy in measurements would be
          > expected, or possible.  Although you can enter fractional amounts
          (e.g.
          > 50.5) if you want.
          >
          >
          > > when i input either 50 or 500, the ControllerGUI hang and
          > > output "invaild input".
          >
          > So, this is pretty odd.  It does bring up a bug that the Controller
          > doesn't currently have a mechanism for noticing when you close a
          window,
          > so after the first warning message, it won't know that you've
          clicked
          > 'ok', and so it won't display any more pop-up warning messages. 
          I'll
          > have to think about that for a bit.  For other windows it's just
          been a
          > minor nuisance of restarting the server on the Aibo, but for GUI
          error
          > messages it's a significant problem.  Regardless, the error message
          is
          > also printed on the console (through serr), so you can look for it
          > there.
          >
          > But I can't get either the 'invalid input' from a number, nor the
          hang
          > to occur.  The invalid input should be followed by actual text of
          what
          > it thinks is invalid.  All you need to type is the number - The
          input is
          > passed directly to the strtod() library function, and I'm just
          looking
          > for an error from that call.  Even when I enter garbage, it just
          > complains and waits for me to try again, hopefully that's all you
          > meant.  (you can still click the "back" button at the bottom of the
          > Controller to get out of the interactive calibration system.)
          >
          > I think this next part might give the clue we need though:
          >
          > > Also, it can not Save Data Set which have the following error
          message
          > >
          > > fs.txt'...a to '/ms/test
          > > Could not open file for savingSoundManager::LoadBuffer() of 11202
          > > bytes
          > > fr.txt'...a to '/ms/test
          > > sr.txt'...open file for savingSaving data to '/ms/test
          > > br.txt'...open file for savingSaving data to '/ms/test
          > > bs.txt'...open file for savingSaving data to '/ms/test
          > > rr.txt'...open file for savingSaving data to '/ms/test
          >
          > Those filenames are wrapping funny - the beginning of the error
          message
          > ("Could not open") is being overwritten by the end of the filename
          > ("xx.txt'...")
          > I'm betting you're entering the filename from the sout console port
          > (10001) or more generally, through a telnet program, which
          uses '\r\n'
          > line endings when you hit enter instead of just '\n'.  And since the
          > '\r' character is not valid for filenames, the error ensues.
          >
          > In the short run, use the "send input" field of the Controller.  In
          the
          > long run, I've checked in an updated Controller.cc to fix the
          problem in
          > the console text processing callback.
          >
          > So that could also explain the problem you were having entering
          values
          > earlier if you were also using the console for that.
          >
          > Hope that fixes things.
          > -ethan
          >
          > Patches:
          > Fixes console input:
          >
          http://cvs.tekkotsu.org/cgi-
          bin/viewcvs.cgi/Tekkotsu/Behaviors/Controller.cc.diff?r1=1.35&r2=1.36
          > Better help text:
          >
          http://cvs.tekkotsu.org/cgi-
          bin/viewcvs.cgi/Tekkotsu/Behaviors/Controls/WalkCalibration.cc.diff?
          r1=1.7&r2=1.8


          Yahoo! Groups Sponsor
          ADVERTISEMENT


          Yahoo! Groups Links
          • To visit your group on the web, go to:

          http://groups.yahoo.com/group/tekkotsu_dev/
            To unsubscribe from this group, send an email to:
          tekkotsu_dev-unsubscribe@yahoogroups.com
            Your use of Yahoo! Groups is subject to the Yahoo! Terms of Service.
        • yuewah99
          Dear ethan, I have been updated the OPENR SDK 1.1.5 and the latest cygwin, mipsel tools. And using the most updated CVS version of Tekkotsu. The error of Walk
          Message 4 of 9 , Mar 3, 2004
            Dear ethan,

            I have been updated the OPENR SDK 1.1.5 and the latest cygwin, mipsel
            tools. And using the most updated CVS version of Tekkotsu. The error
            of Walk Calibration tool before is no longer exist.
            But some problems found,

            "Save" i.e. Save Posture, Walk, DataSet etc. have warning before.
            Now is no problem.

            However, "Load", i.e. Load Posture, is no problem before.
            Now have warning.


            While I use the WalkController to command the AIBO walk in strafe, it
            alway walk in curve instead of straight line. So I want to do my own
            walk calibration. But it is very hard, can I just modify some of the
            default calibration data in order to solve my problem? If not, could
            you explain the fr,sr and br calibration more detail on how to do it
            because i am quite confused by the orginial description of the
            Calibration tool.

            For the sr calibration, "Choose any sideways vel. Tweak forward vel.
            to cancel drift Then choose any rotational vel."
            What do you mean of cancel drift?
            Is it need to maintain heading? And how to do the measure?

            Regards,
            Steven Chan.





            --- In tekkotsu_dev@yahoogroups.com, Ethan Tira-Thompson <ejt@a...>
            wrote:
            > I'm afraid I'm stumped... I thought perhaps this is a windows
            specific
            > thing, but it seems to be working for me, even on windows. I'm 99%
            sure
            > the problem is that you're getting a '\r' character at the end of
            your
            > inputs, but I have no idea how it's getting there from the GUI, or
            why
            > it's not being stripped even if it was there.
            >
            > here's a little tip though - the input processing code can handle
            two
            > inputs - so instead of entering the two measurements separately,
            just
            > put some whitespace between them and it'll use the second word as
            the
            > second input. I'm betting this will let you get around having the
            error
            > pop up.
            >
            > if you figure out what's causing this (and better yet, know how to
            fix
            > it), let me know.
            >
            > sorry,
            > -ethan
            >
            > On Tue, 2004-03-02 at 04:39, yuewah99 wrote:
            >
            > > I am using either Tekkotus and Tekkotsu CVS version.
            > >
            > > In the ControllerGUI, when it request to input the forward
            distance
            > > during the Forward/Strafe Interactive Calibration. I typed "50"
            > > in "Send Input" Textfield and followed by <Enter>.
            > >
            > > The warning message is pop up without any description and the
            > > ControllerGUI is hang.
            > >
            > > console output of ControllerGUI:
            > >
            > > <Error>
            > > >Invalid input:
            > >
            > > However, if i input "!", warning message is pop up. But, i
            input "50"
            > > again without hang.
            > >
            > > For the saving data set problem, i am using the "Send Input"
            > > Textfield in ControllerGUI for the filename instead of system
            console
            > > port 10001. It can save file if i hard code the filename in the
            > > source code.
            > >
            > >
            > > --- In tekkotsu_dev@yahoogroups.com, Ethan Tira-Thompson
            <ejt@a...>
            > > wrote:
            > > > > Who have experience on walk calibration?
            > > >
            > > > Well, I guess that would be me. ;)
            > > >
            > > >
            > > > > What is the dimension of distance to be input while
            > > forward/strafe
            > > > > calibration in Cartesian mode? mm or cm?
            > > >
            > > > All of the distance measurements should be in centimeters. I
            meant
            > > to
            > > > include that in the inline help menu, oops. That will be
            updated
            > > in my
            > > > next checkin.
            > > > Speeds are specified to the underlying walk as
            millimeters/second,
            > > but i
            > > > didn't want to imply that such accuracy in measurements would be
            > > > expected, or possible. Although you can enter fractional
            amounts
            > > (e.g.
            > > > 50.5) if you want.
            > > >
            > > >
            > > > > when i input either 50 or 500, the ControllerGUI hang and
            > > > > output "invaild input".
            > > >
            > > > So, this is pretty odd. It does bring up a bug that the
            Controller
            > > > doesn't currently have a mechanism for noticing when you close
            a
            > > window,
            > > > so after the first warning message, it won't know that you've
            > > clicked
            > > > 'ok', and so it won't display any more pop-up warning
            messages.
            > > I'll
            > > > have to think about that for a bit. For other windows it's
            just
            > > been a
            > > > minor nuisance of restarting the server on the Aibo, but for
            GUI
            > > error
            > > > messages it's a significant problem. Regardless, the error
            message
            > > is
            > > > also printed on the console (through serr), so you can look for
            it
            > > > there.
            > > >
            > > > But I can't get either the 'invalid input' from a number, nor
            the
            > > hang
            > > > to occur. The invalid input should be followed by actual text
            of
            > > what
            > > > it thinks is invalid. All you need to type is the number - The
            > > input is
            > > > passed directly to the strtod() library function, and I'm just
            > > looking
            > > > for an error from that call. Even when I enter garbage, it just
            > > > complains and waits for me to try again, hopefully that's all
            you
            > > > meant. (you can still click the "back" button at the bottom of
            the
            > > > Controller to get out of the interactive calibration system.)
            > > >
            > > > I think this next part might give the clue we need though:
            > > >
            > > > > Also, it can not Save Data Set which have the following error
            > > message
            > > > >
            > > > > fs.txt'...a to '/ms/test
            > > > > Could not open file for savingSoundManager::LoadBuffer() of
            11202
            > > > > bytes
            > > > > fr.txt'...a to '/ms/test
            > > > > sr.txt'...open file for savingSaving data to '/ms/test
            > > > > br.txt'...open file for savingSaving data to '/ms/test
            > > > > bs.txt'...open file for savingSaving data to '/ms/test
            > > > > rr.txt'...open file for savingSaving data to '/ms/test
            > > >
            > > > Those filenames are wrapping funny - the beginning of the error
            > > message
            > > > ("Could not open") is being overwritten by the end of the
            filename
            > > > ("xx.txt'...")
            > > > I'm betting you're entering the filename from the sout console
            port
            > > > (10001) or more generally, through a telnet program, which
            > > uses '\r\n'
            > > > line endings when you hit enter instead of just '\n'. And
            since the
            > > > '\r' character is not valid for filenames, the error ensues.
            > > >
            > > > In the short run, use the "send input" field of the
            Controller. In
            > > the
            > > > long run, I've checked in an updated Controller.cc to fix the
            > > problem in
            > > > the console text processing callback.
            > > >
            > > > So that could also explain the problem you were having entering
            > > values
            > > > earlier if you were also using the console for that.
            > > >
            > > > Hope that fixes things.
            > > > -ethan
            > > >
            > > > Patches:
            > > > Fixes console input:
            > > > http://cvs.tekkotsu.org/cgi-
            > > bin/viewcvs.cgi/Tekkotsu/Behaviors/Controller.cc.diff?
            r1=1.35&r2=1.36
            > > > Better help text:
            > > > http://cvs.tekkotsu.org/cgi-
            > >
            bin/viewcvs.cgi/Tekkotsu/Behaviors/Controls/WalkCalibration.cc.diff?
            > > r1=1.7&r2=1.8
            > >
            > >
            > > Yahoo! Groups Sponsor
            > > ADVERTISEMENT
            > >
            > >
            > >
            > >
            ______________________________________________________________________
            > > Yahoo! Groups Links
            > > * To visit your group on the web, go to:
            > > http://groups.yahoo.com/group/tekkotsu_dev/
            > >
            > > * To unsubscribe from this group, send an email to:
            > > tekkotsu_dev-unsubscribe@yahoogroups.com
            > >
            > > * Your use of Yahoo! Groups is subject to the Yahoo! Terms
            of
            > > Service.
            > >
          • Ethan Tira-Thompson
            I ll have to look into your other questions tomorrow, but in the mean time... ... So, the calibration is based on a least squares solution to a matrix
            Message 5 of 9 , Mar 3, 2004
              I'll have to look into your other questions tomorrow, but in the mean
              time...

              > But it is very hard, can I just modify some of the
              > default calibration data in order to solve my problem?
              So, the calibration is based on a least squares solution to a matrix
              multiplication.
              The idea is to take the 3 intended velocities (forward, strafe, and
              rotate) and multiply this vector through a matrix to obtain 3 command
              velocities to pass to the system such that the actual velocity that
              results matches the desired original commands.
              So the first three columns are straight forward:
              forward strafe rotate

              The problem I discovered is that having a basic matrix doesn't give
              enough flexibility to model the complexities of walking. So, that's
              why there are additional basis functions, which allow various curves to
              be fit as well. The other basis functions, in order, are:
              abs(strafe) , abs(rotate) , gabor , speed squared (forward^2 +
              strafe^2) , forward*rotate , strafe*forward , rotate*strafe

              the last column is always multiplied by a constant 1. So this column
              allows an adjustable offset.

              The last trick is that there are two separate matricies being used -
              one for moving forward, and one for backing up. This might be over
              kill - the idea was that since the aibo isn't symmetrical
              forward/backward, it might have different properties when walking
              backward.

              So anyway, I could get into the details of how the math all works out,
              but hopefully if you play with it for a while you can get an intuitive
              sense of how things fit together. It's just a basic least squares
              problem once you know what basis functions are in use. You can also
              uncomment some of the graph creation calls in WalkCalibration.m to get
              more feedback on the fitting that is being performed.

              But, for a quick example, since you mention it rotates when it should
              strafe, you would want to modify either the 3rd row, 2nd column, or the
              3rd row, 4th column. This value will be multiplied by the intended
              strafe velocity, and added to the commanded rotationality velocity.
              Picking the right one will let you cancel out a rotational drift that
              varies with strafe.

              > If not, could
              > you explain the fr,sr and br calibration more detail on how to do it
              > because i am quite confused by the orginial description of the
              > Calibration tool.
              the idea is to have it walk in an arc - some translational motion while
              rotating.
              Then, by measuring the planar displacement and the angular
              displacement, the software will automatically compute the distance
              traveled itself, so it can record the correct forward/strafe/backward
              speed that was performed.


              > For the sr calibration, "Choose any sideways vel.  Tweak forward vel.
              > to cancel drift Then choose any rotational vel."
              > What do you mean of cancel drift?
              > Is it need to maintain heading? And how to do the measure?
              Exactly - you should adjust the rotational speed so that it maintains
              the same heading - it shouldn't be rotating at all, because your
              measurement for this section will assume there is no rotation performed
              in the real world.

              The whole problem for all of the data sections is that it's practically
              impossible to measure the output of all three parameters at once, so we
              want to vary only two at a time, and cancel the third out so it won't
              affect the measurements.

              hope that helps
              -ethan
            • yuewah99
              Thank you very much. Also, I have figured out that the LoadSave problem is due to the n r again. Since the .mot, .pos file is from yr CVS server, and I
              Message 6 of 9 , Mar 4, 2004
                Thank you very much.

                Also, I have figured out that the LoadSave problem is due to
                the "\n\r" again. Since the .mot, .pos file is from yr CVS server,
                and I found that they are in dos format. So I have converted back to
                unix format, everything is ok now......


                --- In tekkotsu_dev@yahoogroups.com, Ethan Tira-Thompson <ejt@a...>
                wrote:
                > I'll have to look into your other questions tomorrow, but in the
                mean
                > time...
                >
                > > But it is very hard, can I just modify some of the
                > > default calibration data in order to solve my problem?
                > So, the calibration is based on a least squares solution to a
                matrix
                > multiplication.
                > The idea is to take the 3 intended velocities (forward, strafe, and
                > rotate) and multiply this vector through a matrix to obtain 3
                command
                > velocities to pass to the system such that the actual velocity that
                > results matches the desired original commands.
                > So the first three columns are straight forward:
                > forward strafe rotate
                >
                > The problem I discovered is that having a basic matrix doesn't give
                > enough flexibility to model the complexities of walking. So,
                that's
                > why there are additional basis functions, which allow various
                curves to
                > be fit as well. The other basis functions, in order, are:
                > abs(strafe) , abs(rotate) , gabor , speed squared (forward^2 +
                > strafe^2) , forward*rotate , strafe*forward , rotate*strafe
                >
                > the last column is always multiplied by a constant 1. So this
                column
                > allows an adjustable offset.
                >
                > The last trick is that there are two separate matricies being used -

                > one for moving forward, and one for backing up. This might be over
                > kill - the idea was that since the aibo isn't symmetrical
                > forward/backward, it might have different properties when walking
                > backward.
                >
                > So anyway, I could get into the details of how the math all works
                out,
                > but hopefully if you play with it for a while you can get an
                intuitive
                > sense of how things fit together. It's just a basic least squares
                > problem once you know what basis functions are in use. You can
                also
                > uncomment some of the graph creation calls in WalkCalibration.m to
                get
                > more feedback on the fitting that is being performed.
                >
                > But, for a quick example, since you mention it rotates when it
                should
                > strafe, you would want to modify either the 3rd row, 2nd column, or
                the
                > 3rd row, 4th column. This value will be multiplied by the intended
                > strafe velocity, and added to the commanded rotationality
                velocity.
                > Picking the right one will let you cancel out a rotational drift
                that
                > varies with strafe.
                >
                > > If not, could
                > > you explain the fr,sr and br calibration more detail on how to do
                it
                > > because i am quite confused by the orginial description of the
                > > Calibration tool.
                > the idea is to have it walk in an arc - some translational motion
                while
                > rotating.
                > Then, by measuring the planar displacement and the angular
                > displacement, the software will automatically compute the distance
                > traveled itself, so it can record the correct
                forward/strafe/backward
                > speed that was performed.
                >
                >
                > > For the sr calibration, "Choose any sideways vel.?Tweak forward
                vel.
                > > to cancel drift Then choose any rotational vel."
                > > What do you mean of cancel drift?
                > > Is it need to maintain heading? And how to do the measure?
                > Exactly - you should adjust the rotational speed so that it
                maintains
                > the same heading - it shouldn't be rotating at all, because your
                > measurement for this section will assume there is no rotation
                performed
                > in the real world.
                >
                > The whole problem for all of the data sections is that it's
                practically
                > impossible to measure the output of all three parameters at once,
                so we
                > want to vary only two at a time, and cancel the third out so it
                won't
                > affect the measurements.
                >
                > hope that helps
                > -ethan
              • Ethan Tira-Thompson
                Ah! Good job! :) Now we know if anyone else sees this problem what the answer is. The original files are definitely unix format, so I m guessing your cvs
                Message 7 of 9 , Mar 4, 2004
                  Ah! Good job! :) Now we know if anyone else sees this problem what
                  the answer is.

                  The original files are definitely unix format, so I'm guessing your cvs
                  software was trying to be "smart" and convert the line endings for you.
                  There might be an option for that somewhere to keep it from doing that
                  again on future updates...

                  thanks,
                  -ethan


                  On Mar 4, 2004, at 4:52 AM, yuewah99 wrote:

                  > Also, I have figured out that the LoadSave problem is due to
                  > the "\n\r" again. Since the .mot, .pos file is from yr CVS server,
                  > and I found that they are in dos format. So I have converted back to
                  > unix format, everything is ok now......
                • yuewah99
                  You are right, although I enter cvs command in cygwin console, it is actually called the cvsnt, not the cygwin s cvs. ... what ... cvs ... you. ... that ... to
                  Message 8 of 9 , Mar 4, 2004
                    You are right, although I enter cvs command in cygwin console, it is
                    actually called the cvsnt, not the cygwin's cvs.

                    --- In tekkotsu_dev@yahoogroups.com, Ethan Tira-Thompson <ejt@a...>
                    wrote:
                    > Ah! Good job! :) Now we know if anyone else sees this problem
                    what
                    > the answer is.
                    >
                    > The original files are definitely unix format, so I'm guessing your
                    cvs
                    > software was trying to be "smart" and convert the line endings for
                    you.
                    > There might be an option for that somewhere to keep it from doing
                    that
                    > again on future updates...
                    >
                    > thanks,
                    > -ethan
                    >
                    >
                    > On Mar 4, 2004, at 4:52 AM, yuewah99 wrote:
                    >
                    > > Also, I have figured out that the LoadSave problem is due to
                    > > the "\n\r" again. Since the .mot, .pos file is from yr CVS server,
                    > > and I found that they are in dos format. So I have converted back
                    to
                    > > unix format, everything is ok now......
                  Your message has been successfully submitted and would be delivered to recipients shortly.