Object-Centric Knowledge Representations in Hierarchical Planning for Decentralized Environments

Anonymous

Abstract:

This paper focuses on the under-studied problem of automated decentralized AI planning under severe communications and information-sharing restrictions between the planners during planning. Under such conditions, we will argue that traditional logic-based knowledge representations in planning systems are not only brittle and hard to manipulate for reasoning and deliberations but also not suitable for conveying important and maximal information with minimal communications
between the planners or one's observations of another. We describe our proposed representation for object-centric knowledge graphs as a first-class element in our decentralized Hierarchical Task Network (HTN) based decentralized planning framework, called ARCADE. We present a number of preliminary experimental results, enabled by our new formalism.