Multi-agent cognitive planning algorithm based on heuristic search

Abstract

The invention provides a multi-agent cognitive planning algorithm based on heuristic search. According to the algorithm, a main planning algorithm of a multi-agent cognitive planning problem is provided. The main planning algorithm is based on definition of a heuristic function of distance between alternate coverage disjunction form (ACDF) formulae and takes a CAO* algorithm based on AO* and ringdetection as a planner. The multi-agent cognitive planning algorithm has good universality. Any higher-order multi-agent cognitive formula can be processed. The deficient problem that in the existingwork, only a sequence solution can be generated, a disjunction formula cannot be processed and the belief nesting depth is limited is solved. Moreover, compared with a naive forward state space searchmethod in the existing work, the algorithm has the advantages that relatively few nodes are expanded in a search process, a solution of the multi-agent cognitive planning problem can be obtained in relatively reasonable time, some relatively large-scale problems also can be processed, and the relatively good expansibility can be achieved.

Description

technical field [0001] The invention relates to the field of computers, in particular to a multi-agent cognitive planning algorithm based on heuristic search. Background technique [0002] Many artificial intelligence tasks today involve multiple agents communicating with each other and cooperating to complete the task, which leads to two requirements. One is the reasoning requirements for multi-agent knowledge or beliefs. Because in many artificial intelligence tasks, at a certain moment Whether an agent can take a specific action depends on whether the preconditions of the action are satisfied, so the agent needs to reason according to its own knowledge or belief; the second is the evolution of multi-agent knowledge or belief. According to the actions taken by the agent, it may affect the objective environment. Influence may also change the knowledge or belief of the agent, so the knowledge or belief of the agent needs to evolve according to the result of the action. In a...

AI Technical Summary

Problems solved by technology

Since then, Muise et al. have proposed another restricted framework based on two assumptions that do not allow the disjunctive formula and the root agent that takes the entire knowledge base as the belief, but this method also has insufficient expressive ability and acceptable belief nesting. limited depth

[0005] In addition, Huang et al. proposed a general framework for multi-agent cognitive planning based on high-level belief change. This method only considers the common knowledge at the propositional level, and proposes an alternating cover disjunctive paradigm (ACDF) to represent agent knowledge. Library and action, and given the knowledge reasoning and evolution algorithm based on ACDF, and finally based on the PrAO algorithm of breadth-first search to realize the multi-agent cognitive planner MEPK, the limitation of this method lies in its poor scalability, because MEPK is based on the simple The forward state space search algorithm lacks an effective heuristic search method. Therefore, when the scale of test cases and the search space increase, MEPK will expand too many meaningless nodes during the search process, thereby greatly reducing the efficiency of the planner.

Images