Robot skill learning, basis functions, and control regimes

Abstract

A computational, constructive theory of tunable, open loop trajectory skills is presented. A skill is a controller whose outputs achieve any of a family of tasks in a space characterized by n parameters, n>1. Learning consists of a search for the best skill output generation scheme. An interpretation process maps skill outputs into sequences of commands for the plant by using basis functions. It is claimed that appropriate basis functions can speed up the learning process and overcome the limitations of a linear trajectory tuning algorithm. A skill learning algorithm and experiments done with various basis functions for a one-dimensional throwing task are described. Table lookup alternatives and whether modifications might make them feasible in this domain are considered.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>