Interactive Representation Based Minimalist Robot

摘要

Behavior-based robotics have made great advancements in the area of autonomous robotic control. The Subsumption Architecture and its derivatives have proven to be a most effective architecture for autonomous navigation in complex, dynamic environments. Subsumption, however, has no interconnection between the behaviors and no flexibility for the hard-wired behaviors, which drastically limit the ability of subsumption to achieve high level goals. Coupled to this lack of interconnection is the inability of subsumption, or any currently available architecture, to possess system detectable error. This research proposes a fundamentally different type of representation called Interactive Representation. Systems that possess Interactive Representation do not explicitly represent the external environment; these systems represent interactive possibilities with the environment. This development, based only upon interactive possibilities, opens up the ability of a system to possess system detectable error. An architecture was developed based on Interactive Representation, the Interactive Representation Architecture (IRA). Error is based upon a functional definition of task achievement, not upon detection of a sensory fault. There is never any attempt to compare one sensor to another sensor for validation. An error is then defined as anything that causes the circularity of the organization of a system to become broken. This definition of error is implicit instead of explicit, which allows the IRA based systems to truly possess system detectable error and to be able to recover from those errors without the use of an external observer, without resorting to plans, and without increasing the complexity or reducing the speed. Both real and simulated systems utilizing this implicit definition of error were proven to be superior to systems not possessing system detectable error even when there were no faults present.