To describe a planner more formally:.
"A planning system is charged with producing a plan which is a possible solution to a specified problem. The plan produced will be comprised of operator descriptions (templates), provided to the system for each domain of application; in simpler terms, a set of actions. These descriptions are drawn from a fuller set of descriptions which collectively represent the available knowledge of allowable operations in the domain- (1).
A quick planning history.
• General Problem Solver (GPS) by Allen Newell and Herbert Simon in 1957.
• The STRIPS planner, developed by Fikes and Nilsson in 1972.
• The HACKER system, developed by Sussman in 1974.
• NOAH developed in 1975 by Sacerdoti.
• NONLIN developed by Tate in 1977.
• TWEAK developed by Chapman, 1987.
• POP & UC-POP (Soderland & Weld, 1991).
Up until this point there was no accounting for uncertainty in these systems, if you wanted to be able to represent any kind of unresolved condition, then you had to wait until 1992, with the advent of CNLP. CNLP was developed by Peot and Smith in 1992, and was the first planner to include uncertainty; it was able to consider uncertain events without probabilities. The next planning systems to come along, and based on a similar action representation were BURIDAN (Kushmerick et al. 1995) and C-BURIDAN (Draper et al. 1994) these were able to deal with probabilistic planning. This was followed by the CASSANDRA algorithm (Pryor and Collins 1996), which is based on STRIPS and does not consider probabilities. The Weaver architecture (Blythe 1998) used the STRIPS action representation to generate conditional plans and was based upon the PRODIGY (Veloso et al. 1995) classical planning architecture.
Why bother with uncertainty?.
It is necessary to be able to model the effects of uncertainty, as almost all real life problems have areas where there are unknown elements; it is an unavoidable aspect of the world.