This paper looks at how to detect and diagnose mistakes autonomous mobile robots make while navigating through large-scale space using vision. Mistakes are perceptual, cognitive, or motor events that divert one from the intended route. Detection and diagnosis consist of realizing a mistake has occurred, determining what it was, and when it happened. This paper describes an approach that detects mistakes by finding mis-matches between observations and expectations. It diagnoses mistakes by examining knowledge from a variety of sources, including a history of observations and actions. It supports these operations by using symbolic visual information to compare expectations with observations augmented by a priori knowledge. This paper describes MUCKLE, the simulation used to test the approach, and presents experimental results that demonstrate its effectiveness.

This paper describes a navigation system which unites three important capabilities: it enables a mobile robot to avoid obstacles, map the environment, and plan local paths around more complex obstacles while navigating. This system extends and integrates the histogram grid representation and the modified A* algorithm. Our system uses the histogram grid representation and updating method to represent the robot's environment and update a map quickly and reliably with the data from a laser range scanner. Our system also uses the modified A* algorithm to search this grid to plan and replan local paths. The navigation system that we have developed enables a mobile robot to smoothly avoid obstacles while navigating towards a goal at speeds of 0.15 to 0.35 meters per second, speeds which are faster than those typically obtained in similar systems.>

A method for visually detecting moving objects from a moving camera using point correspondences in two orthographic views is described. The method applies a simple structure-from-motion analysis and then identifies those points inconsistent with the interpretation of the scene as a single rigid object. It is effective even when the actual motion parameters cannot be recovered. Demonstrations are presented using point correspondences automatically determined from real image sequences.>