Reinforcement and error feedback differentilly impact motor exploration during locomotion

Abstract

Error-based feedback is typically thought to play the dominant role in learning a new movement pattern, yet reinforcement feedback promises critical impact as well. Reaching studies show that a lack of reinforcement (failure) causes greater variability while positive reinforcement (success) updates the intended motor action towards the last success, leading to greater exploration of the solution space. Here we investigated the roles and interplay of reinforcement and error feedback on motor exploration while walking and the impact on balance due to adjusting to the feedback. Twenty-four healthy young subjects walked on an instrumented treadmill with a 180° virtual reality screen presenting feedback on their step length or step width. Subjects were instructed to match a target step length or width created from their baseline left step length or width. Error feedback displayed their left foot step length or width as a line along with a gray box target length. Reinforcement feedback showed the gray target for each step outside the target range and the target would turn blue when their left foot step length or width was within the target range. Lag-1 autocorrelations, which represent the level of exploratory behavior, were calculated for each condition. A bootstrapped hypothesis test on the paired differences showed that baseline walking and reinforcement feedback led to greater motor exploration than error feedback for step length targets, but not for step width targets. Error feedback yielded corrective behavior regardless of step length or width target conditions. Moreover, lag-1 autocorrelations with reinforcement feedback were no different from baseline walking for step length or width conditions. Analysis of the balance mechanisms showed that step width feedback provides a greater change in balance strategy compared to baseline walking. The reinforcement condition had a larger influence on center of mass on mediolateral step placement than baseline. The center of mass was less predictive of ankle roll for the step width target conditions than at baseline, showing that subjects were relying on feedback to regulate their step dynamics more than their internal sense of center of mass. Our results indicate that the way error and reinforcement feedback are regulated in reaching behavior is generalizable to the regulation of step length. However, maintenance of upright balance leads to reduced exploration for step width targets of error or reinforcement feedback, as well as change in the use of the balance mechanisms.