Subject:
|
Re: Any experience with measuring rotations?
|
Newsgroups:
|
lugnet.robotics.nxt
|
Date:
|
Sun, 21 Jan 2007 00:25:03 GMT
|
Viewed:
|
11405 times
|
| |
|
|
In lugnet.robotics.nxt, Brian Davis wrote:
> In lugnet.robotics.nxt, Rafe Donahue wrote:
>
> > > > So, I guess I need a force sensor, ie, an NXT scale
>
> Well... there's two things here, and Steve solution actually addresses them
> both. Do you actually need a continual series of states, and the "weight" of the
> thing on the end of the beam, or just know if something *is* on the end of the
> beam? I'm either case Steve's solution will work, and it works very well. I used
> a copy of Steve's sensor for a wall-follower and loved the application so much
> I've found it popping up all over (I used it as a "forceless" feeler in my 3T
> machine, and it will actually function just fine to count single soccer balls as
> they roll past on a very modest slope - it can be made that low a force level).
What I need to do compute the average position of a number of items across the
length of the teeter-totter (!). If, on average, they are in the middle, then
there is no rotation and I should get zero. If there are more on the left than
the right, then I want to generate a negative value. Conversely if there are
more on the right.
So, I need to measure, quickly before these things, um, roll off, the average
position of the things. And then I need to reset quick and then do it again.
Of course, I'd like to do it with just one sensor. If I had one sensor for each
potential position (and one computer brick could handle all those inputs) then
that would be fine too; but where's the challenge in that?
But in the end, the real need is for the average position, so any other means to
get there is certainly welcome!
> > I'm thinking I'm going to have to work out a
> > sequence of more than just two shades
>
> I've done this using black, two greys, and white, and using the two rows of
> white 1x1 dots edge-on to get a "calibrated non-linear" scale... but for most
> things, Steve's (again) correct, just use the very slight motion between one
> white and one black object, and scale the motion of the transition to the motion
> you want to detect.
I'm thinking that I'm not quite understanding what Steve's solution looks
like...
> > Can you say Posterior Bayes Estimate?
>
> No. And even after some quick Wiki-ing, I still have no idea. However, I'm
> trying to push the bi-stable flip-flops to new heights of sillyness. Counters,
> clockers, triggerable delay gates.... but for the life of my I can't figure out
> a (practical!) series of async logic gates (well, that *do* anything specific
> anyway).
A posterior Bayes estimate is a probability distribution that results from
integrating a realized likelihood (data) against a prior using Bayes formula.
(Actually the *distribution* comes from that part; the true PBE is found by
doing something like taking the expectation of that distribution...) So, start
with a distribution that is the guess, collect a datum, mix vigorously and bake
at 350 for 45 minutes, and out pops an updated (posterior) distribution. (It
goes faster if you use some type of computer; hint.) This distribution can be
used as the prior for the next iteration. Lather, rinse, repeat, and you have
an adaptive system that uses the data to adjust itself. Some people might call
it an evolutionary scheme or machine learning or something like that.
On a side note, for Christmas my brother got me a binary clock. Somehow, I have
found it very easy to read!
Rafe
|
|
Message is in Reply To:
 | | Re: Any experience with measuring rotations?
|
| (...) Well... there's two things here, and Steve solution actually addresses them both. Do you actually need a continual series of states, and the "weight" of the thing on the end of the beam, or just know if something *is* on the end of the beam? (...) (17 years ago, 20-Jan-07, to lugnet.robotics.nxt)
|
13 Messages in This Thread:
 
 
 
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
