Hybrid approach to path planning in autonomous agents

This paper focuses on the integration of symbolic and sub-symbolic knowledge for the execution of path-planning tasks in autonomous agents. Environmental knowledge is represented through a multilayered architecture whose different abstraction levels are identified by means of meta-knowledge for classification and clustering of distinctive places. The path-planning problem we consider consists in determining the cheapest path for visiting a set of resources in the environment, each resource being expressed as either a cluster or a category of clusters at any abstraction level. Time windows and precedence constraints between resources are taken into account. The algorithm we propose finds a sub-optimal solution to this problem by decomposing it at the different abstraction levels through a divide-et-impera technique.