Motor learning is the shaping of individual sensorimotor capabilities by the physical and social environment. It is based on changes of neural networks of the brain that enable a relatively permanent improvement of performance, even though this may not always be manifest. Motor learning is induced by experience or practice and can take place with the intention to learn but also without; indeed, even without conscious awareness that something has been learned. In this entry, three issues are addressed: (1) the behavioral changes that give evidence of motor learning, (2) the internal representations and mechanisms involved, and (3) the practice conditions that affect motor learning.
Behavioral Changes
Motor learning is associated with a number of behavioral changes. Prominent among them are improvements of speed and accuracy, increasing movement consistency, economy, and automatization.
For many motor skills, higher speed means better performance. Exceptions are skills for which correct temporal patterns are defined, as in playing a musical instrument, or where movements have to be synchronized with environmental events, as in catching balls. The improvement of speed is captured by the power law of practice, which is a good approximation for a variety of skills. According to this empirical law, the time T needed to perform a particular action declines with the number of repetitions N in a way that can be described by a power function, T = kN-α. Across the first repetitions the decline of movement duration is rapid. Later on it becomes progressively smaller, and finally, it can hardly be detected against the background of random variations of performance. Diminishing returns in terms of improvements are a fairly universal characteristic of practice curves not only with temporal measures of performance but also with many other dependent variables. As a consequence, the early stages of practice are generally more rewarding than later stages.
The increase of accuracy in space and/or time is a second major behavioral change. For many tasks, there is a trade-off between speed and accuracy—for example, in movement aimed at a target, known as Fitts’s law. However, motor learning overcomes this trade-off. Typically accuracy increases in concert with speed. Nevertheless, there is an important difference between learning related changes of these two performance characteristics. Accuracy is always defined with respect to a specific target, but speed is mostly “the faster the better”—that is, the goal is maximization (of course, maximization can also be a goal for spatial characteristics in tasks such as throwing a ball as far as possible).
With increasing number of repetitions of a motor skill, performance tends to become more consistent—that is, less variable. This can also represent an increase of accuracy. As an example, consider the task of throwing a ball over a distance of 10 m. The throws will have a mean deviation from the target (called constant error), and they will also vary around the mean (called variable error). A reduction of the constant error typically requires that the learner has information on the error, whereas a reduction of the variable error can result from simple repetitions.
The example of throwing a ball over a certain distance does also illustrate that movement strategies can be adapted to increase accuracy of the outcome, the constant and variable error of throwing distance. The distance covered by a thrown object depends on its initial velocity and the angle of its initial flight path with the horizontal plane. For a certain distance, the required initial velocity is smallest for an initial angle of 45 degrees and for deviations from this angle it increases. Deviations from the strict relation between the initial flight angle and velocity required for the target distance result in different throwing errors depending on the range of flight angles and velocities where they occur. Thus, performance improvement can also result from a proper choice of strategy. More generally, for many motor skills there may be strategies for which variability in movement production has comparatively small effects on the outcome.
During practice of many motor skills, increasing economy of movement production can be experienced. A good example, which unfortunately is somewhat outdated, is the operation of a scythe. This is an exhausting exercise for a beginner, whereas an expert can do it smoothly for a long time. Operation of the scythe requires exact guidance of the instrument. In particular, skills that require high accuracy are initially accompanied by cocontractions. These are concurrent activations of opposing muscles. They do not result in net torques but serve to increase the impedance of the limb and thus make it more resistant against perturbations. In many motor skills, cocontractions decline in the course of practice so that movements become more economical in terms of muscle activity involved and thus in terms of energy consumption.
Finally, motor skills tend to become automatic after sufficient amounts of practice. The “autonomous” phase represents the final stage of a quite popular conceptualization of skill acquisition by Paul Fitts. It follows the initial “cognitive” phase, in which motor patterns are produced with strong cognitive involvement, and the subsequent “associative” phase, in which the components are gradually bound together or associated. Subjectively automaticity comes close to the motor skill running off by itself once it has been started. Functionally, automaticity is inferred from the absence of interference by concurrent cognitive activity. Absence of dual-task interference, however, is not a universal phenomenon with highly practiced motor skills. First, the perturbation of a practiced motor skill by a concurrent cognitive activity depends on the nature of the activity. Second, in the literature, one can find examples that even apparently quite separated cognitive activities such as mental arithmetic interfere with motor skills such as standing (in older persons in particular) or filing of skilled precision mechanics—that is, with skills that seem fully automatic at first glance.
Representations and Mechanisms
In the course of motor learning, the control of movements is changed, which results in better performance. There are underlying changes of internal representations, and there are mechanisms that bring these changes about.
Types of representation, which are acquired or modified during motor learning, are representations of correct movements, of environmental characteristics, and of transformations the motor system has to master. Representations of correct movements have been posited by a number of theories. For example, a core construct in the influential theory of Jack Adams is the perceptual trace. This is a representation of the reafference of the correct movement that is claimed to serve as a reference for closed-loop control. A number of theorists have posited (generalized) motor programs, which are representations of the motor outflow. These have been conceived as prestructured motor commands but also as prototypical force time profiles that can be varied in certain ways so that different variants of movements of a certain type can be produced.
Representations of environmental characteristics have received only little attention in the motor-learning literature. For example, in playing a musical instrument, one can learn which notes are to be played next, or in playing soccer, one can learn what a certain opponent player is likely to do next in a certain situation. Such representations are involved in anticipation, and they can contribute to pretty extensive changes of perception. They also allow to prepare movement sequences in advance and to form chunks of the elements of a sequence, such as in typing, which can then be produced in rapid succession. Sometimes representations of environmental regularities are hard to distinguish from representations of movements. For example, skilled typing involves chunks of finger movements that are produced in rapid succession, but at the same time, these movements reflect the environmental regularity of the letter sequence.
Representations of transformations are generally referred to as internal models. To clarify what is meant by transformations, consider again the example of throwing a ball. Muscles are activated, forces of various muscles are generated and combined, joints are rotated, the hand is moved along a certain path with a certain velocity, and finally the ball flies a certain distance. To produce the proper hand movement and to release the ball at the right time requires a fairly complex internal model of the transformations on the way from muscle activation to ball flight. More precisely, what is required is an inverse model that specifies the proper input to the transformation (muscle activity) that results in the desired output (flight of the ball over 10 m). However, the forward model is also useful in that it allows rapid predictions of the outcome of a movement. In the example, it allows prediction of the success of the throw before the ball has even left the hand.
A popular variant of the notion of internal models is provided by Richard Schmidt’s schema theory, which is actually several years older than the notion of an internal model. According to this theory, a representation called motor schema establishes a link between desired outcomes of an action (e.g., the desired amplitude of a throw) and the parameters of a generalized motor program (GMP) (e.g., the overall force level of a throwing movement or its total duration). One of the major predictions of schema theory is that variable practice should be more beneficial than practice under constant conditions. The reason for this prediction is basically that to learn a relation between variables one has to encounter a range of these variables and not just single values.
Sufficiently accurate (inverse) internal models of the transformations involved in motor control are a prerequisite of open-loop control. In fact, motor learning has been characterized as a progression from closed-loop to open-loop control and also as a progression from the use of visual feedback to the use of proprioceptive feedback. Such progressions may indeed be found for certain motor skills, but there are also motor skills that remain dependent on visual feedback (as well as proprioceptive feedback) even after prolonged practice. Balancing a vertical rod on the tip of the index finger is an example.
Regarding the mechanisms involved in motor learning, repetition effects, error-based corrections, reinforcement, motor resonance, and consolidation are among the important ones. Historically, the existence of repetition-based (or use-dependent) learning without any feedback about the outcome of the action has been doubted, but this is clearly unjustified. Pure repetition can serve to organize visual input (unsupervised learning), for example. Also movements will become faster and less variable. A basis for such changes might be facilitation within the neural networks that are involved in the production of a certain movement and in the perception of the relevant environmental conditions.
Error-based learning and reinforcement learning are sometimes hard to distinguish, and in the early 20th century, the distinction was uncommon. A concept that combines these two types of learning is knowledge of results (KR), which is typically provided after the end of a movement. It can take various formats. For example, it can be evaluative (“good,” “poor”), or it can be informative by way of indicating the precise error (“20 cm too short”).
In the first case there is likely reinforcement, an associated neural event that basically strengthens what has been rewarded (“good”). What has not been reinforced is likely to be changed in the next attempt, and without further information, the change will be more or less random. In the case of informative KR, there is error-based learning in that the next attempt can be modified in a way that compensates the error indicated. In spite of their similarity, these two mechanisms of learning invoke different neural structures.
A currently quite popular mechanism of motor learning is motor resonance. This mechanism is crucial for observational learning. Humans can observe movements produced by other people, and often they can reproduce these movements immediately. The popularity of this kind of learning mechanism has been boosted by the discovery of mirror neurons in monkeys. These neurons are active not only when a certain movement is produced but also when the same movement is observed. In humans, overlapping patterns of activated brain areas have been observed in producing and observing movements and also in producing and imaging them. Thus, observation of a certain movement is likely to activate at least a subset of those neurons that are also involved in its production—the motor system “resonates” in response to the visual input.
A final mechanism is consolidation. Consolidation refers to neural changes that serve to stabilize or even improve what has been practiced before. It is invoked when it comes to an improvement of motor performance after a break or after a night of sleep. It is also invoked when the question is addressed whether the acquisition of a new internal model of a transformation overrides an older internal model or is added to it. In any case, consolidation is the most comfortable mechanism of motor learning in that it does not require any activity of the learner.
Practice Conditions
From the various representations and mechanisms involved in motor learning, a number of principles for the design of practice conditions are obvious. Motor learning requires information to enable error-based learning and evaluative feedback to enable reinforcement learning, variability should facilitate the acquisition of internal models of motor transformations, and sufficient spacing can bring in consolidation and serve to prevent muscle fatigue as well as mental fatigue. Among the less obvious design principles are the avoidance of over-optimization, the use of imagery and observation of the motor skill, and the proper direction of the focus of attention.
In order to optimize practice conditions and to obtain a rapid improvement of performance several measures can be taken. Some of them can produce immediate unintended (and perhaps counterintuitive) effects. For example, error-based learning can be facilitated by sufficiently accurate error information. When errors are presented visually, for example, they can be amplified; when they are presented numerically, they can be given in smaller rather than larger units. The risk of such measures is that the scale of error information exceeds the precision of movement production. In the extreme case, when movements are basically accurate and vary only randomly, the random errors are fed back and learners try to correct them, which is doomed to failure. It only has the effect of increasing variability.
Unintended effects of optimized practice conditions can also become visible after a delay; for example, when the motor skill has to be performed under real-life conditions after augmented feedback has been removed. Performance can become dependent on augmented feedback and break down when augmented feedback is no longer available. In addition, optimization of continuous visual feedback for closed-loop control can impede the acquisition of internal models that are needed for open-loop control. In recent years more and more robots have been designed to support motor (re-)learning, in particular in neuro-rehabilitation. Haptic guidance provided by them results in high levels of performance. It also demonstrates how the correct movement feels. But it prevents active generation and shaping of motor commands as well as active error corrections. Therefore, it can impede rather than facilitate motor learning.
Motor imagery and movement observation can serve to improve motor performance. The reason is that both imagery and observation share neural structures with actual movement production. In a certain way, they are similar—and functionally equivalent—to central processes of overt motor behavior. In general, observational and mental practice are less efficient than physical practice, but combinations of the different types of practice can be superior to physical practice alone.
Observation can offer the opportunity to notice aspects of performance that remain unnoticed when one moves oneself. Imagery offers possibilities of slow motion and the exploration of skill variants that might even be physically impossible. Alternation of periods of observational or mental practice with periods of physical practice allows the combination of the respective advantages of the different procedures.
Finally, during the last several years, a critical role of the distribution of attention for motor learning has been documented. Basically an external focus of attention results in superior learning as compared with an internal focus. The external focus is on movement outcomes—for example, the swing of the golf club—whereas the internal focus is on the moving limbs; that is, the movements of the arms. Most likely the benefits of an external focus of attention are related to the fact that motor control typically starts with a desired outcome (e.g., a throw of a certain distance), and motor commands are selected according to an internal model that has been acquired during motor learning. In the performance of many motor skills awareness of the details of the movements is quite limited, and an internal focus of attention (and thus the attempt to voluntarily control details of muscular contractions) may actually interfere with the required delicate timing of motor commands or other aspects of proper motor outflow.
References:
- Grèzes, J., & Decety, J. (2001). Functional anatomy of execution, mental simulation, observation, and verb generation of actions: A meta-analysis. Human Brain Mapping, 12, 1–19.
- Heathcote, A., Brown, S., & Mewhort, D. J. K. (2000). The power law repealed: The case for an exponential law of practice. Psychonomic Bulletin & Review, 7, 185–207.
- Müller, H., & Sternad, D. (2004). Decomposition of variability in the execution of goal-oriented tasks: Three components of skill improvement. Journal of Experimental Psychology: Human Perception and Performance, 30, 212–233.
- Schmidt, R. A. (1975). A schema theory of discrete motor skill learning. Psychological Review, 82, 225–260.
- Willingham, D. B. (1998). A neuropsychological theory of motor skill learning. Psychological Review, 105,558–584.
- Wolpert, D. M., Diedrichsen, J., & Flanagan, J. R. (2011). Principles of sensorimotor learning. Nature Reviews Neuroscience, 12, 739–751.
- Wulf, G., Shea, C., & Lewthwaite, R. (2010). Motor skill learning and performance: A review of influential factors. Medical Education, 44, 75–84.
See also:
- Sports Psychology
- Motor Development