Optimal control in a value-based automatic behavior composition framework

The behavior composition problem consists in the automatic synthesis of a controller being able to realize a desired target behavior by coordinating a set of available behaviors. Although many frameworks have been proposed for this problem, one of the open issues concerns how to realize the desired target behavior at lower cost when several behaviors are available to perform the same actions. This paper provides an answer to this question by proposing a new framework in which values or costs are assigned to transitions of the available behaviors. Furthermore, controllers are synthesized by using two strategies: one suitable for on-line computation and another for off-line computation. Some experiments show how the two strategies perform on typical data.