Publication at Faculty of Mathematics and Physics |
2017
Abstract
The dawn of autonomous robots brings a question of automated modeling of robot behavior such that the learned robot capabilities can be used to plan robot activities. To bridge the continuous world of sensor readings and control signals with the symbolic world of planning, one needs to identify robot activities as somehow compact behaviors that can be repeated later when a given activity is planned to be performed.
In this paper we focus on identifying activities from a sequence of sensor reading and corresponding control signals by using the methods of machine learning, both supervised and unsupervised. The methods are experimentally evaluated using data from a flying drone.