Next: 4.5 Heuristic Optimizations Up: 4. Memory of Coordinated Behavior Previous: 4.3 Segmenting the Execution Trace

  
4.4 Preparing Execution Trace Segments

Segments are prepared for future re-use before they are stored in memory. The preparation phase makes some representational changes; for example, literals must be replaced by new variables. Also, WAIT-FOR operators are introduced at this time. The most substantial part of the preparation process is to remove actions which are planner-reconstructible.

Two important consequences of removing reconstructible actions are to improve plan quality and to reduce plan-merging effort at communication time. Plan quality is improved since the particular time at which subordinate goals were achieved at runtime may be misleading. Plan-merging effort can be considerable [21] and reducing the level of detail in stored plans reduces this complexity. Another potential benefit of removing planner-reconstructible actions is that it makes the stored plan more easily adapted in the future (at the cost of regenerating the original action if it is needed again).

To compensate for the removal of planner-reconstructible actions, the preparation process augments action descriptions with additional binding information, in the form of new roles and preconditions. This additional information is identified via an annotation process, similar to that of [43], which records the reasons why plan variables took on their final ground value.

The augmentation of action descriptions is better understood with a simple motivating example than with a long explanation. HTO's execution trace from the previous example is:

(<asked L1 to (ON MBOX3 HANDTR2)>
 <waited for (ON MBOX3 HANDTR2)>
 <waited for (ON MBOX3 HANDTR2)>
 <executed (TILT-HANDTR HANDTR2)>
 <executed (PUSH-HANDTR HANDTR2 STREET)>
 <executed (STAND-HANDTR HANDTR2 STREET)>
 <asked L1 to (ON MBOX3 TRUCK1)>
 <waited for (ON MBOX3 TRUCK1)>
 <waited for (ON MBOX3 TRUCK1)>
 <waited for (ON MBOX3 TRUCK1)>
 <waited for (ON MBOX3 TRUCK1)>)

And after preparation:

((SIGN ?ACTOR (ON ?BOX ?HANDTR))
 (STAND-HANDTR ?HANDTR ?FINAL-LOCATION)
 (SIGN ?ACTOR (ON ?BOX ?TRUCK)))

Normally, the SIGN action does not contain a hand-truck role or binding information regarding ?HANDTR. Nonetheless, HTO needs some way to determine an appropriate instantiation for ?HANDTR before the first SIGN can be re-used. This can be accomplished if the action description is augmented with the pair of preconditions (AT ?HANDTR ?STARTING-LOCATION) and (AT ?BOX ?STARTING-LOCATION).


Next: 4.5 Heuristic Optimizations Up: 4. Memory of Coordinated Behavior Previous: 4.3 Segmenting the Execution Trace
Last Update: March 10, 1999 by Andy Garland