<title>Plan-behavior interaction in autonomous navigation</title>

摘要

This paper describes a scheme for plan-behavior interaction in a reactive robot for navigating in an indoor environment. The robot uses very high level plans called " scripts" which resemble the symbolic sequence of instructions that a person would give when asked for directions. The robot is able to execute scripts because of a close correspondence between the script instructions and the four basic abilities hallway traversing (e. g. wall following) landmark recognition place recognition and junction branch selection. The ability invoked in an instruction can often be achieved by more than one of the behaviors associated with that ability. For example ''move down a hallway'' may be implemented using one of the following behaviors: follow wall follow midline follow targeL Thus there is a competition among the movement behaviors in executing the script instruction. Which movement behavior is actually activated at run-time is determined by its priority which is dynamically adjusted depending on the mission goals and the environmental and sensor conditions. Our robot architecture also allows subsumption of the " deliberate" behaviors invoked through a script by the activation of more reactive ones such as collision avoidance in response to changes in the environment. For execution each instruction in a script must be matched to a triple (initiation event behavior termination event). The sequence required by the script is achieved by making the termination event for one instruction