A reactive architecture for planning and executing robot motions with incomplete knowledge

This paper describes a type of contingency-tolerant navigation system for an autonomous mobile robot performing transportation tasks in an office-like environment. A prior model of the environment is available, but this model is incomplete. The system combines a planning component that makes use of the available knowledge to plan a lesser-committed motion plan and a reaction component that pilots the robot according to this plan through the unexpected obstacles detected by the robot's sensors. These two components embody practical answers to several important questions such as what a lesser-committed motion plan is, how it can be represented and used; and how computations should be organized so that the most common situations are handled quickly. The navigation system described has been implemented and a robot called GOFER tested on.<<ETX>>