[ont.events] Adaptive Planning in Complex Uncertain Worlds.

ylfink@water.waterloo.edu (ylfink) (03/28/88)

DEPARTMENT OF COMPUTER SCIENCE
UNIVERSITY OF WATERLOO
SEMINAR ACTIVITIES

RECRUITING/ARTIFICIAL INTELLIGENCE SEMINAR

                    -  Thursday, March 31, 1988

Dr.  R.  James Firby, of Yale University, will speak on
``Adaptive Planning in Complex Uncertain Worlds''.

TIME:                11:30 AM

ROOM:              MC 3005

ABSTRACT

A  robot  acting  in  the real world must be controlled
with a flexible plan.  A plan is usually conceived as a
list  of  primitive  robot  actions  to be executed one
after another. However, in a complex domain a plan must
have   much   more  structure  to  enable  it  to  cope
effectively with the myriad of unpredictable details it
will   encounter   during   execution  and  to  respond
appropriately  to  effector  slips and sudden situation
changes.  Adding structure to a plan involves more than
augmenting  the  primitive plan representation, it also
requires  a  new model of plan execution.  Execution of
one predetermined action after another does not suffice
when  actions  must  be retried or attention shifted to
address unexpected events.

We  propose a plan representation based on program-like
reactive  adaptation  packages,  or  RAPs.   A  plan is
constructed using RAPs and, at execution time, each RAP
decides  which  primitive robot actions are appropriate
for  its  task given the situation encountered.  Within
such   a   system,   execution  monitoring  becomes  an
intrinsic part of the execution algorithm, and the need
for separate replanning on failure disappears.

RAPs  are  more  than just programs to run at execution
time,  however,  they  are  also  hierarchical building
blocks  that  direct plan construction.  Hence, the RAP
representation  is  declarative  enough  that  a  RAP's
expected behavior is evident to the planning machinery.
This  talk presents the RAP planning concept, sketching
out  a  RAP  executor, a simple RAP based planner and a
RAP  representation  that  supports  both execution and
planning.