URBI Forum

I've a problem with the function walk.turn(angle) (implemented since URBI 0.9.8 if I understood well).
I need a function that makes my AIBO turn for a given angle, with a correct accuracy (that's this kind of demand that made the function robot.turn(time) deprecated, right?).
But when I have a chain of 4 sequences of 90? (π/2 for the ones who know ), I don't finish at my starting point (from quite far!!).
I noticed that even if I put a long wait before each sequence, there is a calm sequence followed by a stressed sequence, & so on...
Well, I'm blocked, the function robot.turn(time) doesn't give a better result (even worse, it doesn't turn around a fixed point).
Bad implementation?
Am I the only one to have this problem?
Thanks in advance for your help.
By the way, I found the source code for robot.turn(time) (motion.u), but not for walk.turn(angle).
I use URBI 0.9.8c build295 (thanks jcbaillie).

PS: I leave tomorrow & I come back on Monday 3rd April.
Triple post:
in English on the URBI forum (here)
en fran?ais sur le forum URBI
en fran?ais sur le forum aibo-fr

Dear TCD,

I think this is not a bad implementation. You have to realise that when you are working robots that you will always have some uncertainty and noise in the sensors and motors. A big influence on how well your Aibo approaches a real 360 degrees is the way of walk you use, the kind of floor you use, the amount of wearout in the motors, etc...

You can never expect the Aibo to turn exactly 360 degrees, and i sincerely doubt if 'you' can if you would be blindfolded and asked to turn exactly 360 degrees. I'm almost certain that you will be off by at least a couple of degrees . So the best thing you can do is to do some experiments with the turning algorithm. You can do a series of experiments in which you let the robot turn and you measure both the elapsed angle according to the aibo and the real elapsed angle. With these results you can find out more about any structural and random error in the angles. You could make a regression function to fit your data best and use this as a calibration to compensate the commands that you give to the AIBO. Then you have a calibrated turning algorithm especially suited for you aibo, you kind of floor and you amount of motor wear-out.

I hope this helps. Good luck with it!

Greetings,

Rutger

Hey rutgerv, & thanks for your answer.
I do know about the uncertainty, but in this case I would have expected at least a precision of 10%, that is to say when I ask to turn of 90? (which is not a too small angle to be subject to noise), I would expect a result between 80 & 100? (which is far from being the case, I'm afraid ). (OK, if I was exact, I would have said 81-99?)
I've already done a list of experiments, & unfortunalty the results are too chaotic to make a good regression, & that's exactly why I posted here.
I still think there is a bug in the implementation (see the calm sequence followed by the stressed sequence part of my previous message).
Any other help?
ADD: I think that the result depends a lot on the departure position, see my French post for more informations.