Re: [avr-chat] AVR/Fuzzy logic

[Matthew MacClary]

On Wed, Jun 08, 2005 at 10:21:40AM -0600, Marlin Unruh wrote:
>  Has anyone on this list ever used fuzzy logic on an AVR, or any
>  other MCU?

    Yes, I have used fuzzy logic on AVRs for a couple projects.

>  Is fuzzy logic dead?

    I guess it depends on what country and industry you are talking
about... it has never even been alive in some people's view! The
people and companies who know fuzzy logic seem content to keep using
it to create cool stuff extremely quickly.
    If you are considering using fuzzy logic for a project, here
are some advantages:

1) No math model is required to implement a fuzzy approximator or
   fuzzy controller, so you can just use your intuition about how the
   system should work and then test the controller. This is an
   important feature because mathematically modeling almost anything
   interesting takes a _lot_ of time.

2) Evaluating fuzzy rules can be very efficient, you don't even need
   floating point math let alone exponentiation and integration as a math
   model may require.

    Here are some disadvantages:

1) If you already have a reliable math model, then it may be possible
   to derive optimal control directly from the model using a more
   "conventional" technique.

2) A fuzzy system doesn't automatically have continuous derivatives.


    I do think that it is worth learning about fuzzy logic because it
is a simple method that gives a designer a lot of power. As an
anecdote, I wrote a waypoint following steering controller used by
this autonomous robot in an hour and it worked almost perfectly the
first time we tried it:

http://oregonstate.edu/grandchallenge/

    The robot's computing power consists of exactly 2 16MHz AVRs. Here
is a plot of a different fuzzy controller on that robot:

http://oregonstate.edu/grandchallenge/drupal/node/20

-Matt

---------------------------------
Where did the universe come from?