Abstract:

The ability to coordinate behavior is at the heart of successful social interactions such as in cooperating to perform tasks or imparting and receiving information. Learning to execute movements in strategic timing relationships with others therefore is a key conceptual and a practical aspect of acquisition of skills useful for everyday functioning. Using the empirical/theoretical framework of coordination dynamics, we present a hybrid human-machine system called Virtual Partner Interaction (VPI) to explore a methodology for skill learning. In VPI, a subject coordinates rhythmic effector actions with that of a computational model of him/herself tasked to execute similar or opposing actions. The interaction is bidirectional: while viewing the virtual partner’s actions through a monitor, subject’s movements are digitized and fed to a computer which integrates the equations of motion that in turn animates the display. The bidirectional coupling between the subject and the virtual partner is governed by a modified Haken-Kelso-Bunz equation of behavioral coordination. In learning using VPI, we build on the Kelso-Zanone model in which learning is defined operationally as the creation of an attractor corresponding to the target behavior to be learnt. In classical normative approach the performance of the task to be learned is often measured against a group “standard” and is assessed how far from the goal individuals are at various points in time. With the Kelso-Zanone model, the learned behavior (a new attractor) is evaluated within the subject’s own landscape of pre-existing stable behaviors. Used in conjunction with this learning model, as a dynamical system, VPI affords one to describe the path from the initial to the current condition at various time-points once a goal is prescribed. This allows one to monitor the stability of the subject’s behavior and the slope of learning. Previous work on VPI showed the existence of emergent coordination patterns (under certain parameter conditions), i.e. coordination behavior patterns present during bidirectional interaction but which may vanish in unidirectional (VP-to-Subject or Subject-to-VP) cases. Additionally, VPI coupling has been shown to be modulated by the character of the movement trajectory as well as by the visual features of the stimulus. Here, we explore the possibility that the creation of the attractor for the pattern to be learned may be facilitated by (1) the presence of the bidirectional interaction, and (2) modulating the coupling via VP parameters. Simply put, the virtual partner acts as a surrogate teacher whose level of participation at all levels of the learning process may be adjusted at will. For example, during the learning process, it may demonstrate the pattern to be learned (high VP-to-subject coupling values), “induce” the subject to execute the pattern but without the VP “doing all the work” (medium coupling values) or simply test the subject’s performance (VP acts as a passive rhythmic stimulus). Additionally, by modifying the animated image the last case may be used to analyze a form of transfer: if a learned pattern can be performed with a generic stimulus (e.g. moving dot on a monitor). Data from a previous work on VPI will be presented which supports the feasibility of the methodology.

Keywords:

virtual partner interaction, coordination, dynamic clamp, learning.