Enhancing Motor Learning in Cycling Tasks: The Role of Model Predictive Control and Training Sequence

Abstract

We evaluated the impact of Model Predictive Control (MPC) robotic-assisted compared to unassisted training, and the effect of alternating between MPC-assisted and unassisted training sequences on motor learning of a complex bicycle steering task. This task involved participants steering and collecting virtual stars displayed on a screen riding a steer-by-wire bicycle on a treadmill. Ten participants were split into two groups, alternating between MPC-assisted and unassisted training. Tasks’ motor skills were quantified by the distance to stars and its standard deviation (SD), while motor performance was determined by mean absolute and SD of steering rate across three evaluation time points: Baseline, Mid-Training, and Post-Training. The repeated-measures ANOVA indicated a significant improvement in task skill (SD of distance from stars) and steering performance (mean absolute and SD of steering rate) and an interaction effect of Group x Time Point on mean absolute and SD of steering rate. The group who initially trained without MPC exhibited a notable decrease in average and variation of steering rate, implying an advantage in starting training unassisted. Our findings suggest that the strategy of starting the training unassisted could stimulate an internal focus (concentrating on one’s own body movements) and intrinsic skill perception, which forms a basis for later integrating MPC assistance to further refining the gained motor skills. Such a sequential training approach may be beneficial in motor skill acquisition of complex dynamics tasks. MPC assistance could be advantageous for individuals with diminished internal model and skill perception, such as those with balance impairments, potentially allowing them to rely less on their impaired sensorimotor abilities.