This paper studies the complexity of two different algorithms proposed as extensions of A* for multiobjective search: MOA* and NAMOA*. It is known that, for any given problem, NAMOA* requires the consideration of no more alternatives than MOA* when provided with the same heuristic information.
In this paper we show that, in fact, expansions performed by MOA* can be many more than those demanded by the problem, and hence than those performed by NAMOA*. More specifically, we show a sequence of problems whose size grows linearly such that the number of expansions performed by NAMOA* grows also linearly, but the number of expansions performed by MOA* grows exponentially. Therefore, there are problems where NAMOA* performs exponentially better than MOA*.
[1]
Alberto Martelli,et al.
On the Complexity of Admissible Search Algorithms
,
1977,
Artif. Intell..
[2]
Chelsea C. White,et al.
Multiobjective A*
,
1991,
JACM.
[3]
Dan Suciu,et al.
Journal of the ACM
,
2006
.
[4]
Lawrence Mandow,et al.
A New Approach to Multiobjective A* Search
,
2005,
IJCAI.
[5]
Lawrence Mandow,et al.
Comparison of Heuristics in Multiobjective A* Search
,
2005,
CAEPIA.
[6]
Pierre Hansen,et al.
Bicriterion Path Problems
,
1980
.