Freezing and freeing of degrees of freedom in joint action learning

In daily life, we often encounter situations in which we have to coordinate our actions with others to achieve a common goal. These actions are also defined as joint actions. In this study we investigated how a multi-agent system learns to acquire control in a novel joint action task. To this end, we designed a task in which agents had to coordinate their actions so as to control a ball rolling on a long, hand-held beam. Participants' task was to roll the ball as fast and accurately as possible back-and-forth between two indicated targets on the beam, by manually adjusting the inclination angle of the beam. In the joint action version of this task, two participants each hold a different beam extremity. In a solo action version, the participant holds one extremity while the other is attached to a static support. The experiment consisted of two practice sessions that each comprised 15 two-min trials. One group of 12 participants first performed a solo action session of the task and then a joint action session (Group S/J), while another group of 12 participants started with a joint action session, followed by a solo action session (Group J/S). While performance increased over practice in all sessions, we found that in the joint action task dyads without prior solo task experience (Group J/S) adopted a sequential pattern of interpersonal coordination by freezing their motion whenever the other agent moved. In contrast, dyads that had received prior practice in the solo task setting (Group S/J) demonstrated less freezing and more complementary motion during the joint action performance. Lastly, we found that initial practice as a dyad in the joint action task did not result in a significant improvement of a subsequent solo action performance. We concluded that multi-agent motor learning in a novel joint action task is characterized by the initial freezing of task-relevant degrees of freedom, while individual training in a constrained setting can stimulate the freeing of these DFs during subsequent joint action performance.

Automated summary

This study investigates how a multi-agent system learns control in a joint action task, finding that dyads with prior solo task experience exhibit less freezing and more complementary motion, while those without prior solo task experience adopt a sequential pattern of interpersonal coordination.