Localisation and world modelling: An architectural perspective

Abstract

Autonomous robot world modelling is a "chicken-and-egg" problem: position estimation needs a model of the world, whereas world modelling needs the robot position. Most of the works dealing with this issue propose holistic solutions under an algorithmic perspective by neglecting software architecture issues. This results in huge and monolithic pieces of software where implementation details reify strategic decisions. An architectural approach founded on separation of concerns may help to break the loop. Localisation and modelling, acting on different time scales, are mostly independent of each other. Sometimes synchronisation is required. Whenever needed, an external strategy tunes the relative rates of the two activities. The paper introduces rationale, design, and implementation of such a system which relies on Real-Time Performers, a software architecture providing suitable architectural abstractions to observe and control the system's temporal behaviour.