Experiments in Adaptive State Space Robotics (1989)

Andrew Moore

Tags

Applications, Memory-based Learning

Abstract

This paper introduces some of the issues addressed by adaptive state-space robotics, a theory of coordinated perception and action. Central to the theory is the assumption that the behaviour of a system (robot or organism) is determined by a state-machine, in which the set of states defines the space of possible actions of the system. Of particular interest is how the control space can be incrementally defined as the system observes how various control decisions affect it own behaviour. The aim is to find out what can be done without using a system model and distal trajectory plans. For implementation, we have not been as much concerned with the "neural net" approach, as with the more general problem of efficiently constructing nonlinear multivariate mappings by self-adaptation. We describe some simulation studies and implemented robot control experiments. We also discuss a number of possible relationships between this work and behaviour in animals.

Full text

Download (application/pdf, 5.4 MB)

Approximate BibTeX Entry

@inproceedings{clocksin-experiments,
    Month = {April},
    Year = {1989},
    Pages = {115-125},
    Publisher = {Morgan Kaufmann},
    Address = {2929 Campus Drive, San Mateo, CA 94403},
    Booktitle = {Proceedings of the 7th AISB Conference, Brighton},
    Editor = {Anthony G Cohn},
    Institution = {University of Sussex},
    Author = { Andrew Moore},
    Title = {Experiments in Adaptive State Space Robotics}
}