Beyond task-space exploration: On the role of variance for motor control and learning

• Published in: Frontiers in psychology 2022-07, Vol.13, p.935273-935273
• Main Authors: Hossner, Ernst-Joachim, Zahno, Stephan
• Format: Article
• Language: English
• Subjects: adaptation; learning mechanisms; meta-learning; motor variance; Psychology; sensorimotor learning; variability
• ISSN: 1664-1078; 1664-1078
• DOI: 10.3389/fpsyg.2022.935273

This conceptual analysis on the role of variance for motor control and learning should be taken as a call to: (a) overcome the classic motor-action controversy by identifying converging lines and mutual synergies in the explanation of motor behavior phenomena, and (b) design more empirical research on low-level operational aspects of motor behavior rather than on high-level theoretical terms. Throughout the paper, claim (a) is exemplified by deploying the well-accepted task-space landscape metaphor. This approach provides an illustration not only of a dynamical sensorimotor system but also of a structure of internal forward models, as they are used in more cognitively rooted frameworks such as the theory of optimal feedback control. Claim (b) is put into practice by, mainly theoretically, substantiating a number of predictions for the role of variance in motor control and learning that can be derived from a convergent perspective. From this standpoint, it becomes obvious that variance is neither generally "good" nor generally "bad" for sensorimotor learning. Rather, the predictions derived suggest that specific forms of variance cause specific changes on permanent performance. In this endeavor, Newell's concept of task-space exploration is identified as a fundamental learning mechanism. Beyond, we highlight further predictions regarding the optimal use of variance for learning from a converging view. These predictions regard, on the one hand, additional learning mechanisms based on the task-space landscape metaphor-namely task-space formation, task-space differentiation and task-space (de-)composition-and, on the other hand, mechanisms of meta-learning that refer to handling noise as well as learning-to-learn and learning-to-adapt. Due to the character of a conceptual-analysis paper, we grant ourselves the right to be highly speculative on some issues. Thus, we would like readers to see our call mainly as an effort to stimulate both a meta-theoretical discussion on chances for convergence between classically separated lines of thought and, on an empirical level, future research on the role of variance in motor control and learning.