A semantic-based flexible framework for automatic behavior composition

Abstract

The behavior composition problem consists in the synthesis of a controller that coordinates a set of available behaviors with the aim of realizing a desired target behavior. This problem suffers from a serious drawback because, for many instances, the target behavior cannot be achieved. In that case, the user is responsible for bringing changes until the problem becomes solvable. This inherent limitation is due to the fact that an exact match is required between the actions initiated by the user and those offered by the available behaviors. The main question is, how can a controller select a suitable behavior such that the functionality of one of its available actions is close to that currently requested by the desired target behavior? This paper provides an answer to this question by associating semantics to actions based on expectations and exploiting a semantic similarity function between actions. Similar actions that have different names, but comparable tasks, are considered equivalent. This renders the behavior composition framework more flexible so that it can fit with semantic environments.