Variable kinetic constraints in robotic assistance enhance learning of a complex motor skill

Abstract

Abstract

Robotic assistance can improve the learning of complex motor skills and support rehabilitation. However, most active robotic devices developed to date provide assistance based on movement trajectories (kinematics) rather than the underlying limb dynamics (kinetics) that generate movement. To our knowledge, no study has directly compared, within a single task, how kinematics and kinetics should be integrated to enhance motor learning. The present study investigated how a complex torque–motion profile involving the right arm can be learned by manipulating torque error margins within an admittance-controlled robotic device. The participants practiced under four acquisition conditions—Constant 2 N·m, Constant 12 N·m, increasing margin (Margin+), and decreasing margin (Margin−)—in which the width of the error margin on the required torque was manipulated to induce different levels of kinetic constraints on practice. On Day 1, all the participants first completed a pretest without visual feedback, followed by 6 blocks of practice. During this acquisition phase, group-specific error margins were applied, and participants received both concurrent and terminal visual feedback of their produced torque. On Day 2, a posttest without visual feedback was performed to assess retention (identical as pretest). The results revealed that the variable-margin groups (Margin + and Margin−) led to greater improvements in kinetic and kinematic profile tracking and more stable retention than the constant-margin groups (Constant 2 N·m and Constant 12 N·m). These findings suggest that variable-margin practice enhances learning by fostering increased exploration of the task space while emphasizing both kinetic and kinematic aspects of task performance. Whereas constant margins appear to stabilize a single movement solution, variable constraints may promote a more generalizable motor control strategy, possibly by refining the predictive mapping between motor commands and their sensory consequences that contribute to updating internal models. Such insights could optimize the design of robotic rehabilitation protocols aimed at fostering durable motor recovery and relearning.