RE: Forward Error Correction and under-runs at speed
Does the error depend on the distance moved? If the error is, say, 10mm at
the highest acceleration for movements of, say, 5 cells and 10 cells, it
would demonstrate that the problem is at the start of acceleration. If the
error is 5mm at 5 cells and 10mm at 10 cells, it would show that the error
is proportional to distance.
My bet would be that it is independent of distance and is due to slippage
at the start of the move.
"Back heavy" and "front heavy" are relative. I made a quick guess based on
The slider is twice as far from the wheel contact points as the c of g is
above these points.
The desired acceleration is 1g.
At one g acceleration, a front slider takes no weight if the c of g is half
way between the wheels and the slider. The static weight on the slider is
HALF the mouse weight. The initial acceleration would need to be about 0.4g
and ramp up as weight is transferred to the wheels. I doubt that you went as
far as this level of imbalance.
I have been contemplating this problem for some time and hope to present a
paper at MINOS about it. There seems to be a trade off between corner speed
and acceleration. I have a suspicion that the "best" solution is an inverted
pendulum that actively balances itself. This can guarantee 100% of the
weight on the wheels at all times but the software would be non-trivial.
[mailto:firstname.lastname@example.org]On Behalf Of Peter Harrison
Sent: 22 November 2007 21:31
Subject: Re: Forward Error Correction and under-runs at speed
I ran the tests again.
Four squares forward at 0.5, 1.5 and 2.5m/s/s
Top speed was limited to 2 m/s so it will never have been reached.
The mouse was run back-heavy, balanced and front-heavy. The front and
back-heavy runs were about the same level of imbalance.
At the lowest acceleration, the actual distance travelled is measured.
at the middle acceleration setting, the mouse stops short by about 4mm.
At the highest acceleration, the mouse stops short by about 10mm.
The same results were obtained with all three setups although I did not
measure the distance in each case to the nearest anything - just marked
lines on the maze floor.
Adding more weight to the rear does, however, seem to make things worse.
While the front and rear skid are similar (cheap domed drawing pins),
the rear skid is more worn than the front one and may well present a
I shall have to perform all the tests in reverse to be sure. Maybe
tomorrow as everyone has gone to bed early.
Acceleration and velocity are stored as 8.8 fixed point values, distance
as a 24.8 fixed point value. Simple additions accumulate acceleration to
speed and speed to distance. The mouse is stepper driven with 16x
microstepping giving a resolution on the ground of about 0.05mm per
step. Distance travelled can be determined from a counter attached to
each wheel pulse generator or from the accumulated speed. In either case
the result comes out the same. The mouse thinks it has gone so far but
reality falls a little short with greater acceleration.