next up previous
Next: Planner Assumption 1: Is Up: Interpretation of Results and Previous: Problem Assumption 2: How

Problem Assumption 3: Does Performance Depend on PDDL Requirements Features?

The planners did not perform quite as advertised or expected given some problem features. This discrepancy could have many possible causes: problems incorrectly specified, planners with less sensitivity than thought, solutions not being correct, etc. For example, many of the problems in the benchmark set were not designed for the competitions or even intended to be widely used and so may not have been specified carefully enough.
Recommendation 8: When problems are contributed to the benchmark set, developers should verify that the requirements stated in the description of each problem correctly reflect the subset of features needed. Planner evaluators should then use only those problems that match a planner's capabilities.
Depending on the cause, the results can be skewed, e.g., a planner may be unfairly maligned for being unable to solve a problem that it was specifically designed not to solve. The above recommendation addresses gaps in the specification of the problem set, but some mismatches between the capabilities specifiable in PDDL and those that planners possess remain.
Recommendation 9: Planner developers should develop a vocabulary for their planner's capabilities, as in the PDDL flags, and specify the expected capabilities in the planner's distribution.

next up previous
Next: Planner Assumption 1: Is Up: Interpretation of Results and Previous: Problem Assumption 2: How
©2002 AI Access Foundation and Morgan Kaufmann Publishers. All rights reserved.