This paper reports the lessons of a number of demonstrator projects to build low-cost mobile robots for unconstrained environments. It is a discursive discussion of the problems in the design of a system architecture for such a mobile robot with goal directness that bases its actions on a dynamically maintained world model. The architecture described is based on two parallel hierarchies of algorithmic perception and action/motion tasks that the author proposes be supervised with a production system written in OPS5.
The author approaches path planning as a graph search problem in a network of discrete decision points with the robot’s perceptual system verifying the accessibility of the decision points. (These were learned initially by the robot as it explored the environment by traveling along the walls.)
The author also indicates a particular general-purpose vehicle-level protocol developed during the projects. He highlights the necessity to have a composite world model that is capable of fusing abstracted data from a number of primarily ultrasonic sensors.
The paper is an introduction to the more technical writings of the author, which also make up the bulk of the references. It is unfortunate that the author did not also direct his readers to the work of others in the field.