Date: 02 Jun 2005

Learning and adaptivity: Enhancing reactive behaviour architectures in real-world interaction systems

* Final gross prices may vary according to local VAT.

Get Access

Abstract

The success of the behaviour-based approach to designing robot control structures has largely been a result of the bottom-up development of a number of fast, tightly coupled control processes. These are specifically designed for a particular agent-environment situation. The onus is on the designers to provide the agent with a suitable response repertoire using their own knowledge of the proposed agent-environment interaction. A need for learning and adaptivity to be built into robot systems from the lowest levels is identified. An enhancement to basic reactive robot control architectures is proposed that enables processes to learn and subsequently adapt their input-output mapping. This results in both a local and global increase in robustness as well as a simplification of the design process. An implementation of the proposed mechanism is demonstrated in a real-world situated system: the control of an active laser scanning sensor.

Funded by the Engineering and Physical Sciences Research Council and Sira Ltd.