🚀 go-pugleaf

I also did not see the original post, but this is a topic which is important
to the work of my group. 

Real Time: 

I define "real time" to mean providing a result within a fixed delay. 
We form our systems as a layered set of servo loops, each working in a
more abstract "space". (Unlike NASREM, we allow loops at teh same level 
to communicate). (See "Coordinating Action and Perception in a Surveillance
Robot" in IJCAI 87 or in IEEE Expert, Winter 87. )

I take AI to mean "symbolic reasoning". Such techniques sit just above
our algorithmic "actions" and are also organised as a servo loops. 
Symbolic Reasoning systems are notoriosly slow; Worse it is difficult or
impossible to predict the cycle time. 

The solution is to let the symbolic reasoning system operate as a 
separate process. Under UNIX and VRTX this process communicates with 
other processes by sockets. 

Memory: 

The other half of the problem is that most real time systems have limited
memory, while most symbolic reasoning systems run on LISP. Many LISP's 
are notoriuosly BIG. For years we have used OPS5 to build our supervisors. 
We have been using Dan Miranker's (U Texas) C based ops5. This allowed
us to squeeze the MITHRA Mobile RObot supervisor onto a 2 megabyte memory
on a dedicated 68020. 

> Seems to me that one could do everything in 'C' by using CLIPS [C Language
> Integrated Production System].....

We obtained a copy of CLIPS a few weeks ago. It's GREAT! Its integration
into a C based system seems direct. It also fixes a number
of complaints about ops. And, the keywords are more natural. 
We are likely to do all future real time symbolic control work in clips. 

    James L. Crowley, LIFIA (IMAG), I. N. P. Grenoble



-- 
------------------------------------------------------------------------
James L. Crowley  LIFIA (IMAG), 46 av. Felix Viallet, 38031 Grenoble  FRANCE
  tel: (33) 76 57 46 55 -- fax : 76 57 46 02   email : jlc@lifia.imag.fr
------------------------------------------------------------------------