Todorov E, Shadmehr R, Bizzi E (1997) Accelerated learning of a difficult motor task through training in a virtual environment. Journal of Motor Behaviour, 29:147-158.

Abstract A number of techniques can be used to augment the feedback that a subject usually receives during training on a motor task (e.g., from video taping to computer enhancement of the environment). While some forms of augmented feedback have been shown to enhance performance on isolated isometric tasks during training, subjects have sometimes not been able to perform as well as controls in the "real-world" task when the feedback has been removed. Indeed, for realistic, non-isometric motor tasks, improved skill acquisition due to augmented feedback has not been demonstrated. Here we report on a set of experiments performed with a system that we designed for teaching a difficult multi-joint movement in a table tennis environment. The system was a fairly realistic computer animation of the environment and included paddles for the teacher and subject, as well as a virtual ball. Each subject attempted to learn a difficult shot by matching the pattern of movements of the expert teacher. Augmented feedback focused the attention of the subject on a minimum set of movement details that were most relevant to the task, and was presented in a form that required least perceptual processing. Effectiveness of training was measured by the performance in the real task. We found that subjects that received the virtual environment training performed significantly better than subjects that received a comparable amount of real-task practice or coaching. Kinematic analysis indicated that practice with the expert's trajectory served as a basis for performance on the real-world task, and that the movements executed after training were subject-specific modifications of the expert's trajectory. However, practice with this trajectory alone was not sufficient for transfer to the real task: when a critical component of the virtual environment was removed, subjects showed no transfer to the real task.