Reactive control of a mobile robot requires to map robot's perceptions into actions by means of a strategy that is goal-oriented. To explicitly encode this strategy is not an easy task. Specifically, it cannot be assumed that users of future service robots will be able to perform this kind of low level robot programming. What is sought for is a method that allows to intuitively program the robot without requiring knowledge about robot's hardware, its sensor system, robot's perceptions and actions, and the actual relationship between them. In this paper, the acquisition of basic mobility skills from human demonstrations is advocated to be such a method. Based on data obtained from manual operation of the robot only, a description of the demonstrated skill is generated, that is operational with respect to the robot. Since human demonstrations are neither optimal nor do they cover the real state space sufficiently, skill adaptation and extension are also considered.
Machine Learning, Robotics, Man-Machine Systems, Programming Support
All articles published by Studies in Informatics and Control (SIC) are published in accordance with the terms and conditions of the Creative Commons Attribution License (CC BY-NC-SA 4.0).