Generation and classification of individual behaviours for virtual players control in motor coordination tasks

The interaction of robots or physical/virtual avatars with humans will be increasingly common in a number of different scenarios intended to improve the quality of human life. For example, in the domain of healthcare, they can support therapists aiding patients in need of motor rehabilitation and so on. In this context, a fundamental control problem is to synthesize strategies to make the artificial agents interact with humans in a “natural” and human-like fashion. This is particularly relevant when artificial agents are required to coordinate their motion with humans performing joint tasks. It has been shown that for rehabilitation purposes, virtual agents in motor coordination tasks must exhibit certain kinematic properties (or Individual Motor Signature) that are characteristic of human motion. In this paper we discuss a method based on the use of Markov chain to generate artificial individual motor signatures that can be used to provide online reference signals for the control of virtual agents. The methodology is also used to classify and identify individual motor signatures belonging to individuals affected by social disorders.