Implicit motor adaptation is driven by motor performance prediction error rather than sensory prediction error

The motor system uses error signals arising during action to adapt future actions. A key question is how are useful error signals constructed from sensory information about action outcomes? Previous studies have suggested two forms of error that affect adaptation: sensory prediction error (SPE), the difference between the actual and predicted sensory information about an action, and motor performance error (MPE), the difference between actual performance and the task goal. These two errors differ in that MPE is dependent on the task goal, but SPE is not, whereas SPE is dependent on internal prediction, but MPE is not. It is widely believed that SPE drives implicit adaptation. Here, we find that, in line with the idea of SPE-driven implicit learning, prediction plays a key role in driving this adaptation, however SPE-driven learning itself fails to explain key features of the implicit adaptive response. Instead we find that motor performance prediction error (MPPE), the difference between predicted performance and the task goal consistently explains the implicit adaptive responses seen during a series of experimental manipulations that dissociate SPE, MPE, and MPPE by perturbing actions and goals both separately and in combination.