Abstract
Motor adaptation under naturalistic perturbations remains understudied. Most paradigms use simplified tasks constrained to a single dimension solution space—such as guiding a rotated cursor onto a target. In contrast, many real-world motor behaviors involve adapting to external forces in tasks with high-dimensional solution spaces, such as throwing objects under wind or water resistance. To address this gap, we developed a VR paradigm in which participants adapted to a constant lateral water current while performing a throwing task.
Participants threw a virtual ball toward four distributed targets—two upstream, two downstream—across alternating training phases with and without the perturbation. Targets were presented in a blocked design (one of each type per phase), allowing us to assess generalization to novel target sets. A physics model was used to derive each target’s solution space, defined as the set of launch angle–speed combinations that result in a hit. Upstream targets exhibited narrower solution spaces, imposing performance constraints, while downstream targets permitted a broader range of successful strategies.
To evaluate learning, we compared mean endpoint errors between early and late training trials for each target. Only the upstream targets—associated with the largest initial error and narrowest solution space—showed a reduction in endpoint error with training. Likewise, throws to these targets appeared to benefit the most from training, suggesting some degree of generalization. These results suggest that naturalistic motor adaptation may emerge more readily under constrained conditions, and that prior experience with similar constraints can facilitate partial generalization to new contexts.
