Human Motor Learning Dynamics in High-dimensional Tasks

TL;DR

The paper advances a high-dimensional motor-learning model (HML) that leverages motor synergies via $C=W\Phi$ to reduce learning complexity and couples fast forward and slow inverse learning with a perception layer to capture continuous visual feedback. It establishes convergence properties and validates the model against data from a target-capture task with 19 finger joints, showing close replication of human learning dynamics. The authors further explore trade-offs—exploration-exploitation, speed-accuracy, satisficing, and flexibility-performance—demonstrating how parameter tuning shapes learning and performance. These insights hold potential for designing adaptive training and assistive-control strategies in complex motor tasks and rehabilitation contexts.

Abstract

Conventional approaches to enhancing movement coordination, such as providing instructions and visual feedback, are often inadequate in complex motor tasks with multiple degrees of freedom (DoFs). To effectively address coordination deficits in such complex motor systems, it becomes imperative to develop interventions grounded in a model of human motor learning; however, modeling such learning processes is challenging due to the large DoFs. In this paper, we present a computational motor learning model that leverages the concept of motor synergies to extract low-dimensional learning representations in the high-dimensional motor space and the internal model theory of motor control to capture both fast and slow motor learning processes. We establish the model's convergence properties and validate it using data from a target capture game played by human participants. We study the influence of model parameters on several motor learning trade-offs such as speed-accuracy, exploration-exploitation, satisficing, and flexibility-performance, and show that the human motor learning system tunes these parameters to optimize learning and various output performance metrics.

Figures (9)

• Figure 1: Performance measures across subjects: Temporal evolution of (a) reaching error, and (b) straightness of trajectory for subjects (red) and the respective fitted HML model (blue) across trials.
• Figure 2: Cursor Trajectories: Cursor trajectory data from human experiments (a), (c) and the fitted model (b), (d). As learning progresses through the $8$ sessions, the trajectories become closer to a straight line between targets, which the proposed HML model also captures. (e) shows the evolution of forward model error for the fitted model as a function of trials.
• Figure 3: Comparing HML model with Ref pierella2019dynamics model: Comparing the errors in RE curve fitting from the model in Ref pierella2019dynamics to the HML model shows that the model in Ref pierella2019dynamics is not as accurate as HML model in capturing the RE for this motor learning task.
• Figure 4: Effort variation with $\eta$: Distribution of driving and exploratory effort (averaged across $128$ Monte Carlo runs) with means and $95\%$ confidence bounds across trials as $\eta$ is varied around its fitted value $3.1742$. While driving effort increases, exploratory effort decreases initially, and both plateau past the fitted $\eta$ value. One-tailed paired t-tests over the effort values across trials reveal this plateauing effect at a significance level of p$<0.001$.
• Figure 5: Speed and accuracy variation with $k_P$: Across trial distribution (averaged over $128$ Monte Carlo runs) of speed and accuracy with means and $95\%$ confidence bounds as $k_P$ is varied around its fitted value $1.3098$. Accuracy is highest around the fitted value (p$<0.001$) and past that speed increases while accuracy decreases.

Theorems & Definitions (9)

• Definition 1: Sufficiently Rich Signal
• Remark 1
• Lemma 1: Stability of the Reduced System \ref{['SI:BLS']}
• proof
• Lemma 2: Stability of the Boundary Layer System \ref{['SI:RS']}
• proof
• Theorem 1: Stability of the HML Model \ref{['SI:sspm_u']}
• proof
• Remark 2