Error Detection and Error Recovery in Two Dimensional Topological Navigation

In this paper we describe error detection and error recovery methods applicable to large scale unstructured environmental navigation. We relax our prior assumption of error-free following of topological landmarks; the navigator is “permitted” to make mistakes during its journey. The error detection method involves the navigator observing its immediate environmental surroundings, and checking for one of several types of qualitative disparities. The error recovery method is based on a simple fixed set of movements, which are triggered by the navigator’s local observations. Alternately described, this work characterizes those environments in which robust topological navigation is possible, including those landmarks which, literally, “can’t be missed”. These methods have been implemented on our qualitative environmental navigation system consisting of a camera mounted IBM 7575 robot arm.