Observation Information Reduction In Nondeterministic Domain On Model Checking

The uncertain planning is a hot issue in the study of artificial intelligence. Themethod on model checking has great advantage in dealing with many unceitainplanning problems. In the completely observability(or partially observability), it isone of the important part for planning the reachability objective.In the realistic planning areas, accroding to the current information of observedvariables, the controller should determine the next action which the agent willperform to execute a planning solution. Some observation information is not used inthe controller execute the planning solution. It must spend a cost to obtain theseobservation information, so in the premise of the controller execute a planningsolution correctly, how can reduce to obtain the information is a very necessary thingto think.This paper have researched on the reduction of observation information, and takethe observation reduction divided into two kinds of situations, the one is that reducethe infromation under the condition of known planning solution, and the another isthar reduce the infromation under the condition of unknown planning solution. At first,under the condition of known planning solution, this paper have detailed that the stepsof the reduction of observation information, where find a needed state-pair set andthen reduce the observed variables set by it, and designed ORSCP and ORSCPM twoalgorithms to reduce information for strong circle planning, and the main differencebetween these two algorithms is the method for finding the need to differentiate states.In the same domain, the effect of these two algorithms will be different, so select theappropriate algorithm according to the actual situation.At present, in the multi-agent planning domain, most research have focused onthe deterministic domain, where the outcome of an action is certain, but the researchis few in nondeterministic domain. In nondeterministic multi-agent domain under thecondition of unknown planning solution, this paper designs a ORMAP algorithmwhich can find a collaborative planning. Firstly, this algorithm takes the state layeredby the hierarchical thinking to find out the possible state of conflict. Secondly, searchthe collaborative planning by priority to minimize the cost of backtracking, wherereduce the observation information. At last, get a collaborative planning which thecontroller execute with a little information in numerous meet the conditions of thecooperative planning solutions. Through the experiments and analysis shows that thealgorithm efficiency is higher.