Motor Learning Driven by Sensory Prediction Errors is Insensitive to Task Performance Feedback

Abstract

Accurate motor behavior relies on our ability to refine movements based on errors. While sensory prediction errors (SPEs), mismatches between expected and actual sensory feedback, predominantly drive such adaptation, task performance errors (TPEs), or failures in achieving movement goals, also appear to contribute. However, whether and how TPEs interact with SPEs to shape net learning, remains controversial. This controversy stems from difficulties in experimentally decorrelating these errors, ambiguity related to possible interpretations of task instructions, and inconsistencies between theory and computational models. To try and resolve this issue, we employed variants of an "error-clamp" adaptation paradigm across four reaching experiments (N = 144). Addressing the ambiguity of whether or not the TPE is indeed ignored in standard error-clamp designs as assumed in theoretical (but not computational) models, Experiment 1 explicitly manipulated TPE magnitude by shifting the endpoint feedback location while holding SPE constant. We found that learning was uninfluenced by TPE size. Experiment 2 assumed that the TPE is in fact disregarded under clamp instructions. To then study SPE-TPE interactions, we induced TPEs of varying magnitudes by shifting the target location ("target jump") while always clamping feedback to the original target location. Here, instructions to reach the new target also induced an SPE. Crucially, learning driven by this SPE was again unaffected by TPE magnitude, a result validated by two additional experiments. Our findings consistently demonstrate that SPE-mediated learning remains impervious to variations in task performance feedback, and point to a distinction in learning mechanisms triggered by these two error signals.