Reinforcement learning in motor control

How do we learn motor skills such as reaching, walking, swimming, or riding a bicycle? There is a large literature on motor skill acquisition which is full of controversies (for an introduction to human motor control, see Schmidt and Lee, 1999), but there is general agreement that motor learning requires the learner, human or not, to receive response-produced feedback through various senses providing information about performance. Careful consideration of the nature of the feedback used in learning is important for understanding the role of reinforcement learning in motor control (see REINFORCEMENT LEARNING). One function of feedback is to guide the performance of movements. This is the kind of feedback with which we are familiar from control theory, where it is the basis of servo control, although its role in guiding animal movement is more complex. Another function of feedback is to provide information useful for improving subsequent movement. Feedback having this function has been called learning feedback . Note that this functional distinction between feedback for control and for learning does not mean that the signals or channels serving these functions need to be different.