Flexible planning by integrating multilevel reasoning

The successful application of a general-purpose planning system in real-world complex application domains depends on the extent to which the activities of plan formation, plan execution, monitoring and failure handling can be flexibly interleaved. This paper presents a multilevel reasoning system (called MRG) where low-level functionalities (e.g. navigation, vision and speech understanding) are flexibly combined with higher-level reasoning capabilities through a set of specialized subreasoners. At the top of the architecture there is a high-level reasoning system, implemented using a programming language based on the notion of tactics. MRG is currently used as the core of an intelligent robot, able to navigate within an unpredictable environment and to exchange information with human users.