People form metacognitive representations of their own abilities across a range of tasks. How these representations are influenced by errors during learning is poorly understood. Here we ask how metacognitive confidence judgments of performance during motor learning are shaped by the learner’s recent history of errors. Across four motor learning experiments, our computational modeling approach demonstrated that people’s confidence judgments are best explained by a recency-weighted averaging of visually observed errors. Moreover, in the formation of these confidence estimates, people appear to re-weight observed motor errors according to a subjective cost function. Confidence judgments were adaptive, incorporating recent motor errors in a manner that was sensitive to the volatility of the learning environment, integrating a shallower history when the environment was more volatile. Finally, confidence tracked motor errors in the context of both implicit and explicit motor learning, but only showed evidence of influencing behavior in the latter. Our study thus provides a novel descriptive model that successfully approximates the dynamics of metacognitive judgments during motor learning.