Requests for primitive actions are placed on a queue. A queue manager is maintained that consumes such requests and determines which agent resource should be used to execute the request. For example, the queue manager may determine which hand should be used to execute a grasp based on following criteria: which hand is empty, which hand is closer to the object to be grasped and also potentially which hand the agent favors for object manipulation (e.g., is the agent right or left handed?).
A queue of action requests facilitates the animation of sensory procedures. Such procedures may be executed parallel or in addition to motor actions. For example, task related attention is supported in our human model. The queue manager automatically invokes the appropriate attentional behavior for each type of task on the queue. While the agent is walking to a goal, for example, he will look at the goal and occasionally glance at his feet(when a memory uncertainty threshold is exceeded or when the agent is in close proximity to an obstacle). When grasping an object, the queue manager invokes an attentional process that determines which sites are relevant for the grasp and directs attention appropriately. Our aim is to direct attention as autonomously as possible and based on analysis of task mix. Since a queue allows lookahead of downstream tasks, we allow the potential to interleave or anticipate where attention may be directed.