The objective of our research is to establish a universal architecture for effecting an autonomous control logic for land, space, air, and underwater platforms. Although the operational constraints of each platform are diverse, we perceive the general task to be the execution of a mission in an unstructured environment. The task requires reflexive and deliberative mechanisms that sometime trade off problems and sometime combine to solve a single problem. We propose a hierarchical time-ordered architecture for integrating these mechanisms.
Integration RequirementsAssume that a platform is unmanned and incapable of communication with an external agent. The platform is equipped with a sensor suite, an effector suite, a mission, and (optionally) a priori information about itself and the environment in which it is operating. The task is to respond appropriately to unanticipated events. Unanticipated events refer to environmental and platform anomalies which are unplanned, e.g. the appearance of an unexpected obstacle, the mechanical failure of a piece of hardware, or the inability to control vehicle dynamics within a predetermined range.Typical categories of responses are do nothing, modify mission, or abort mission. The components of the task that require integration are reflexive behavior (RB) and deliberative behavior (DB). For example, suppose a platform has planned a path to a goal (DB) and comes upon an unexpected object. If time permits, the platform should replan a path around the obstacle and proceed to the goal (DB). If time does not permit replanning of a path, the platform should stray from the object (RB) and place itself in position to replan a path to the goal (DB). A third course of action is to stray from the object and replan a path to the goal simultaneously (RB/DB). Other examples of behavior that require both RB and DB are discussed in Georgeff and Lansky [10], Schoppers [23], and Dean and Boddy [5]. The goal of our research is to understand how RB and DB trade off problems, how they combine to solve a single problem, and how decisions for one or the other evolve in dynamic environments.
Architecture ComponentsThe distinction between RB and DB is grounded in (i) the time available to respond to an event, (ii) information abstraction, and (iii) the ability to project into the future. Essentially, DB requires more time, more abstract knowledge, and more ability to think about the future than RB. These properties of the task suggest a time-ordered architecture (TOA) (cf. also Simmons and Chappell [24], and Crowley [4]). See Figure 1. z < COORDINATION y SENSORS EFFECTORS rn 0 O N Z Figure 1: The Time-Ordered Architecture (adapted from Hebert et al [12], page 2)The TOA is a multiple-loop mechanism in which control problems are partitioned into classes based on required response time and information abstraction. Each class of problems resides on a single level of the TOA, and a problemsolving technique on a single level operates on a uniform abstraction of data (e.g. signals, signs, or symbols). RB emer...