When discrepancies between planned and actual movements arise due to environmental changes, humans adjust movement parameters to achieve task goals. While motor adaptation has been extensively studied, the mechanisms involved in redundant movement parameters remain unclear. Split-belt treadmill adaptation, where each belt moves at a different speed, is an example of this phenomenon. Such adaptation initially induces gait asymmetry, which diminishes over time. Previous studies have postulated step length asymmetry as the target function;​ however, recent evidence challenges this assumption, leaving the target function undefined. This study investigates the target function by analyzing step parameter asymmetry using the goal-equivalent manifold and generalization predictability. The goal-equivalent manifold assesses whether adaptation is close to optimal in minimizing step parameter asymmetry, while generalization predictability reflects adaptation effects across different contexts, indicating potential target functions. We propose that step velocity asymmetry, rather than step length asymmetry, serves as the target function in split-belt treadmill adaptation. This framework facilitates the prediction and interpretation of both the learning process and the transfer of learning effects from trained to untrained conditions. In addition, it explains the overadaptation of step length asymmetry and the achievement of energy-efficient gait after adaptation. Therefore, we propose that step velocity asymmetry is the primary target function in split-belt treadmill adaptation.