Re: NAO Robot : is Ada there too ?

["Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de>]

On Mon, 13 Jun 2011 17:50:13 +0200, Yannick Duch�ne (Hibou57) wrote:

> Le Mon, 13 Jun 2011 11:39:43 +0200, Dmitry A. Kazakov  
> <mailbox@dmitry-kazakov.de> a �crit:
>> I doubt that Prolog can be useful for the AI deployed in a robot. There  
>> is not much inference it would do, more problems like path finding,  
>> which are not logical. Further, even considering logical problems you  
>> rather need fuzzy inference than Prolog's stuff. And certainly Prolog is absolutely
>> useless for image processing and control.

> I started some time ago (not yet finished) a Prolog interpreter (actually  
> in SML, the final implementation is planned to be Ada), in the purpose to  
> solve some things I don't like with typical Prolog interpreters like  
> SWI-Prolog (mostly with handling of infinite recursions, which I handle as  
> something which is interpretable, and the so called Cut, which I dropped).  
> As I worked on this, I can say with reasonable belief, that unification  
> may use weighting, instead of exact match.

The problem with all approaches to uncertainty be it fuzzy or
probabilistic, is that you cannot create loss-less inference. It does not
exist. A fuzzy engine has to track all possible alternatives defuzzifying
at the output using some heuristics to reduce the problem. Same is true for
probabilistic approaches with the difference that heuristics are much
weaker.

> any coherent system has logic.

Yes, any system has an interpretation within standard logic.

(which does not really help. E.g. does not create you an Ada compiler by
mere logical conclusions from the ARM and the input program (:-))

> Then, any fuzzy measurements, in AI,  
> should probably be finally stored as a kind of symbolic data finally  
> (abstracted down).

Uncertain measurements are continuous. Yes you do split truth levels /
probabilities into intervals, etc, but they leak into each other.

> This would just require dynamic creation of symbols. I  
> honestly do not see any other way to do: complex data needs to be  
> abstracted down, because the system must incorporate more and more  
> knowledge, without growing linearly (compared to the quantity of data it  
> has handled). Finally, this ends to reduce complex data to symbols.

Complex data, yes, an Ada program is complex data. No other way to build a
compiler than by using symbols (for each program one, I guess? (:-))

IMO, it would be easier to write a custom inference engine from scratch
than to bend Prolog to do things it cannot.

P.S. If you decide to do something, I have a compiler from Fuzzy Control
Language written in Ada.

-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de