Introduction

Abstract

Robots must operate in a world which is inherently uncertain. This uncertainty arises in the perception and modeling of environments, in the motion of manipulators and objects, and in the planning and execution of tasks. In present-day industrial robotics, this uncertainty is represented implicitly in programming and planning, requiring errors in operation to be small and well understand. Robust task execution is accomplished by increasing sensor and manipulator accuracy, using high tolerance and well known objects, and supplying parts in predetermined positions and orientations. This works well in situations where the environment, tasks and motions are all completely determined in advance. However, as robot systems are increasingly required to operate in more complex domains, where the environment and tasks are uncertain or unknown, then this implicit allowance for uncertainty will be insufficient to ensure efficient operation.