Motor control and action - summary of chapter 8 of Cognitive Psychology by Gilhooly, K & Lyddy, F, M

Cognitive Psychology
Chapter 8
Motor control and action

How our body achieves our goals.
The description of motor control and action in three parts:

• How we use our motor system (the components of the central and peripheral nervous system along with the muscles, joints and bones that enable movement) to produce movement. No matter the motor activity, it is being coordinated by the nervous system and implemented by muscles.
• Strives to understand how units of motor behavior can be strung together.
• How the motor system interconnects with other psychological functions

Motor control

How body movements are planned by the brain and performed by the body.

Movement between targets could be described with a two-component process of motor control.

• The impulse phase initiated the movement and was planned in advance of the start.
  In essence the brain would calculate in advance what limbs to move and how they should move and this plan in the from of a motor command would be sent from the brain out to the body.
• A control phase where vision is key to controlling the accuracy of the final endpoint position.

Degrees of freedom: of a joint are the number of ways it can move.
When performing a task, the joints do not need to all move in all possible ways.
This provides us with great versatility in performing actions in changing situations.

The computational problem of how to plan a movement out of the multitude of alternatives.
(Like the inverse problem in vision 3D → 2D)
Given all the possible factors for achieving a goal, we can see that the motor planning system is controlled with a difficult task to plan what body parts will move with what motion.
The effortlessness in with we act to achieve our goal shows that our brain has worked out an efficient strategy for producing movements.

Theories of movement planning

Three approaches:

• Equilibrium point hypothesis
  A theory of motor control that emphasizes how the problem of control can be simplified by taking into account muscle properties.
  Emphasizes the special relationship between the brain and the muscles.
  Muscles exert different forces depending on how much they are stretched.
  Any stable posture requires the setting of various control parameters for muscle activation to achieve stability.
  Moving from one posture to another can be achieved by simply resetting these parameters so that the spring-like properties of the muscle move you into the next posture.
  This planning exploits the spring-like properties of the muscles to simplify what the brain must control to move the body.
• Dynamical systems
  Emphasizes interaction between the body and the environment and uses special mathematics that describe how a system’s behavior changes over time.
  Related to ecological theories of psychology.
  Motor control as a process of self-organization between an animal and its environment.
  Dynamical systems is a branch of mathematics that includes rules which describe the evolution of the state of a system over time.
  Like: the tendency that when we move two limbs together, there is a tendency for them to exhibit mirror symmetric movements.
• Optimal control theory
  Provides a framework for implementing principles that optimally satisfy some criterion.
  It does not focus on constrains of the body, but instead views motor control as the evolutionary or developmental result of a nervous system that tries to optimize organizational principles.
  In optimal control, the problem of planning a movement is solved by using an optimization principle to define the best movement. It includes planning a movement to be the smoothest motion between two points, planning the least amount of torque-change at the joints, or planning the least amount of spatial errors in task achievement.
  Torque: a measure form physics that measures rotational force.
  The characteristics of coordination are determined by the movement structure imposed by the optimization principle used.

The optimal control theory is an advanced form of simple feedback mechanisms.
After the original impulse phase the control phase is entered and sensory information is available to evaluate whether the goal has been achieved and suitably modify the movement using feedback.
Forward model: used to predict the relationship between actions and their consequences. Given a motor command the forward model predicts the resulting behavior of the body and the world. The forward model obtains these predictions form simulating the effects of our commands. Mental access t to the prediction of a motor command is necessary because we need to be able to move quickly in complex ways. Predictions of the sensory consequences of our motor commands are available faster than the sensory feedback resulting from the motor command.
These does not seem to be time enough for feedback to be involved in highly skilled performance.

The challenge of the motor system is to deal with the world as it is now based on sensory information that is a tenth of a second old and with motor commands that will take effect in muscles a tenth a second in the future.

The process of optimal control theory is cycliar, with motor commands being sent out of a control policy, and the result of the motor coming back to the control policy in the form of an estimate of the state that reflects how the motor command has changed things.
The cycle:

1. Control policy
   Takes as input the current state estimate and outputs a motor command. It provides a set of rules that determine what to do given a particular goal and state estimate.
2. Motor command.
   Output form the control policy and contains the information about how the body is supposed to move.
3. Noise
   Physiological noise is due to imperfect neural transmission along the pathway from brain to body
4. Forward model
   Takes as input the motor command and outputs a prediction of the sensory consequences of the motor command.
5. Body and the world
   Takes as input the motor command that has been degraded by noise and produces an action that changes the state of the body and typically also the world. This creates new sensory information.
6. Sensory information
   The changes to the body and world create sensory information
7. Noise
   The sensory information is also corrupted by physiological noise arising form imperfect sensing and neural transmission. This leads to uncertainty in estimating the state of the body and the world.
8. Sensory integration
   Takes as input all the sensory information as well as the prediction of the forward model and outputs an estimate of the current state of the system
9. State estimate
   Provides an internal representation of what is the current state of the body and world and this is input to the control policy.

The optimal control theory describes aspects of motor control that are not fixed by physiology or the relationship between organism and environment.
The state estimate upon which a movement is planned is heavily influence by the forward model and the reliability of sensory information.

When creating a plan for movement, we incorporate our knowledge of the uncertainty of the visual information and the motor apparatus to plan movements that will gain us the highest reward.
One can take the different ‘boxes’ in the optimal control model and map these functions onto the known functions of different brain areas involve with motor control.

Forward model → cerebellum
Sensory integration → parietal cortex
Control policy → basal ganglia

Producing complex actions

Movement goals an how these lead to complex sequences of actions.

Action sequences

Humans are continually active and this activity has a complex temporal structure that appears only in animals with a highly developed brain.

Associative chain theory.
A behaviorist theory that explains how sequences of action arise from linking together associations between individual action components.
The end of one particular action is associated with stimulating the start of the next action in sequence.
This can be an effective method for simple and limited sequences, but it has difficulty with general sequences.

There were two key ideas from speech production that were used to advance cognitive models of serial planning.
These two ideas are

• The pattern of errors we make when we speak
• How the production of different speech sounds are coordinated to produce fluent speech.

It seems as if before the sentence begins all the words are somehow available and ready to make a particular grammatical structure.
Errors occur by misplacing words in the structure rather than in purely sequential errors of what goes after what.
The production of speech involves several inter-related but somewhat independent neurological systems.
Different articulators of the vocal tract are critical at some times of an utterance and not critical at other times.
When an articulator is not critical it is able to prepare for upcoming sounds to be produced as long as it does not interfere with intelligibility of the current sound being produced.
This leads to a phenomenon known as co-articulation where the target sound is being articulated at the same time that future sounds are being prepared.
Control of speech articulators are best modeled as arising from the interaction of separate mechanisms governed within a hierarchy of constraints.

Hierarchical models of action production

Different mechanisms could work simultaneously in parallel to create sequences.

• Test-operate-test-exit (TOTE) unit.
  Once selected, a TOTE unit would continuously test whether a condition was met and then exit once the condition was satisfied.
  (For example find the matching sock out a pile of socks)
  This architecture also allowed TOTE units to call other TOTE units, thus permitting a hierarchical structure to be used to produce a sequence of actions.
• Hierarchies of control elements which activated other control elements at the levels below.
  Each node of the hierarchy corresponds to a particular action schema.
  The action schemas and their position in the hierarchy may directly onto the sequence of actions that need to take place for the action to be achieved.

To produce an action sequence it must be possible to traverse the hierarchy in a manner that activates currently desired units while suppressing currently undesired units.
Models of how hierarchical structures can be used to produce sequences have developed from theories of recurrent networks.
Recurrent networks: a type of artificial neural network with connections between units arranged so to obtain a cycle of activation. This design allows a temporal context to be designed into the computation. They can be designed to control the timing of operations.
Interactive activation: a term used to describe the pattern of network activity generated by excitatory and inhibitory interactions of feature detectors and object representations. When one unit of a hierarchy is selected for activation, other units at the same level of hierarchy are inhibited.
This facilitates the selected action schema to complete.

Brain damage and action production

The frontal cortex is responsible for action planning and it is thought that, the coordination of action is set in an organized manner across the anatomy of the frontal brain.
This organization can be seen as a hierarchy with high level control of planning performed in anterior portions of the frontal cortex, and as one goes from anterior frontal cortex towards motor cortex the brain areas are involved in increasingly elemental aspects of control.

Damage to the frontal cortex is often diffuse across several regions of frontal cortex and leads to conditions such as Dysexecutive syndrome, and action disorganization syndrome, where patients make frequent errors in producing action sequences.
These slips include such slips of action as:

• Insertions
  (Like entering the room and turning on the light even though it is day),
• Confusions
  (Like putting shampoo on a toothbrush),
• Perseveration
  (Repeatedly picking up and putting down a toothbrush),
• Omissions
  (Leaving a key ingredient out when preparing food and not noticing it till dinner).

Actions slips are common in typical individuals.

Action disorganization syndrome fits into a broader family of movement disorders.
Apraxia: a neurological condition typically resulting from brain damage where a person loses the ability to perform activities that they are physically able and willing to do.
Can arise from patterns of brain damage to the frontal and parietal lobe, basal ganglia and the nerve fibers connection these regions.

Idemotor apraxia: an inability to pantomine tool use and gesture when verbally instructed to do so. They can do it when the correct stimulus is there.

Action representation and perception

Theories of action representation

Cognitive sandwich: the view that perception and action are like slices of bread that surround cognition as the filling of a sandwich.

Cognitive representations of actions intermingle with representations of both perception and action.

Historical perspectives

There is a long tradition known as ideomotor theory that intimately connects perception to action.
Relates how thinking about the results of an action can give rise to producing the action.
Sensory codes are translated to motor codes by cognitive meanings.

Common codes for action perception and production

Common coding:
A theory of perception and action production which holds that both production and perception share certain representations of actions in the world.
Addressed the problem of how sensory codes can be internally related to motor codes.
There is a layer of representation that includes event codes and action codes. In this extra layer, aspects of event coding overlap with those of action coding.

One predicted consequence of common coding is that because common code is a resource for both perceiving and producing action there would be interference between perception and production when they both tried to access the same resource at the same time.

Another consequence: there will be properties of action that are common to both perception and production. The same regularity between speed and geometry captures both motor and visual processes.
When viewing a target that obeyed the power law there was extensive brain activation not only in visual areas, but also areas related to motor production.

A final result from common coding:
The hypothesis that if we observed action of another is similar to how an observer would perform the action themselves, then there will be greater overlap of the common codes.
Viewing your own performance of an action would lead to a more effective activation of common code and this would lead to individuals excelling at identifying their own actions.

Common coding can explain how performance changes when perception and action compete for the same resources.
It can also explain how regularities between perception and action would exist and why we can recognize our own actions better than those of another.

Mirror mechanisms and action observation

A neural mechanism to unite perception and actions was provided by the discovery of mirror neurons.
Mirror neurons: neurons with the special property that they represent both the sensory aspects of perceiving actions as well as motor aspects of how to produce the action.

From the discovery of mirror neurons neuroscience research proceeded in two directions:

• Further explore the monkey’s brain for other evidence of mirror neurons.
  This resulted in mirror neurons being found in the parietal cortex (that respond to doing and seeing) as well as in the premotor cortex (that respond to doing and hearing).
  The fact that mirror neurons were found in two regions of the brain and involved the coding of two forms of sensory information suggest that this is a general information processing strategy, rather than a mechanism with a limited scope.
• To explore whether mirror neurons would be found in the human brain.

It is argued that the crucial function mirror neurons provide is access to the goal of the movement.

One proposed function of the human mirror neurons is imitation and learning motor actions from visual models.
The function of the MNS in imitation learning is not limited to representing the goal of the observed action, but also to represent the motor primitives.
Motor primitives: the basis set of elemental movements that serve as building blocks for an animal’s repertoire of movements.
The MNS is activated more strongly during important information. Passive observation induced weaker activations than observation in order to imitate.
In addition to the MNS, the prefrontal cortex was also activated during action observation and motor preparation.

This implies that a (mirroring) mechanism of automatic perception-action matching alone is insufficient to account for imitation learning, instead higher-order supervisory operations associated with the prefrontal cortex are involved which most likely engage in the manipulation and restructuring of the elementary motor representations provided by the MNS.

Embodied cognition

How does embodied cognition relate to motor control and action representation.
In the embodied view of cognition, perception and action are intimately connected.

In the most radical views of embodiment, the connection between perception and action is so tight and connected with the environment, there is little need for the kinds of abstract symbolic representations often used to explain cognition.

Less radical view of embodiment hold that for simple actions, motor information is incorporated directly into representations. However, for complex actions both perceptual and motor information are combined in a flexible manner.

Connections between perception and action that demonstrate embodiment are present in common coding and mirror neurons.
Visual representation is not independent of the action response.

There are several consequences of recent work on embodied cognition:

• To elevate our appreciation of the importance of the body in cognition, and to include it as an important component to understanding cognitive performance.
  The unique structure of the body are essential to how the brain comes about its control strategy.
• To recognize the importance of the environment in cognition.
  We organize our environment to reduce cognitive load.
• How the body plays a central role in metaphor.
  Communication of complex information relies on an embodied representation of the concepts we wish to express.

Gesture

An important topic that illustrates how the body and action are related to cognitive processes is gesture.

Even alone, gesture can convey clear messages.

There are a variety of different types of gestures.

• Deictic gestures
  Are pointing movements done in order to draw attention to a location or thing in the world.
• Beat gestures
  Are baton-like movements that do not appear to have a direct meaning, but instead are used in tight synchrony with speech to accent important aspects of the information being conveyed.
• Metaphoric gestures
  Exploit the structure of a metaphor to understand one thing in terms of another by using the spatial structure and timing of a movement to relate to concepts being communicated.
• Iconic gestures
  Depict physical properties of the object of reference.

These definitions of gestures are not exclusive since a single gesture might have a component of several different types of gestures.

Gestures result because ideas are being simulated in terms of perceptual and motor properties, regardless of whether the idea is about something physically spatial or only metaphorically spatial.