Oh, we are back here again! What a live topic! Anyone think this makes sense as an answer? I originally wrote it as a proposal for a journal article that was rejected and I need to write the full article at some point!
Predictive (or feedforward) theories of motor control have replaced simple feedback theories. Yet
the limitations of feedforward theories have been recognized and „hybrid‟ models have attempted
to reintegrate feedback, but within a secondary role. We critique hybrid models and propose that
the field would progress by „going full circle‟ to make feedback control primary. Sensory feedback
is critical because it corrects inaccurate predictions. One contemporary feedback theory
(Perceptual Control Theory; PCT) can manage delays in neural signaling, circumvent limits in
sensory feedback, and learn to optimize control. We identify four original components of PCT as
alternative ways of implementing feedforward processes to enhance feedback control over four
different, parallel timescales. We discuss the implications for a unified understanding of control.
‘Going Full Circle’: Is Control More Important Than Prediction? Warren Mansell 7/2/12
My article proposes that the „hybrid‟ models of control overemphasize prediction at the expense of
control and should be superseded by returning „full circle‟ to a feedback theory of control that
subsumes feedforward processes within its architecture.
I plan to start the article with accessible examples. At the turn of the 20th century, John Dewey
proposed that “the motor response determines the stimulus just as truly as the sensory stimulus
determines movement” - while our environment can trigger changes in our behavior, it is also the
case that our behavior is affecting the environment and we sense these effects as ‘feedback’. Take
the example of chasing a moving target during hunting – a vital skill for any predator’s survival.
Processing current sensory feedback is critical to control. The prey may change direction at any
moment, as may the angle and terrain of the ground beneath the predator’s feet. Moreover, as the
predator shifts its speed and angle to compensate, the visual image of the prey it perceives on its
retina will change instantly. The brain of the predator must process current sensory and motor
information in such a way that it regularly, and efficiently, reaches its target – or ultimately it dies.
Early 20th century engineers, recognizing the importance of feedback control, developed
technologies we take for granted today, such as thermostats, amplifiers, flight control systems, and
industrial and medical devices - some of which have an accuracy of up to four parts per million.
Within psychology, feedback models peaked with the development of ‘cybernetics’ (Ashby, 1952;
Wiener, 1948). Within a few decades, however, the pendulum had swung towards the
development of feedforward theories of control (Oosting & Dickerson, 1987). Feedforward theories
are varied, yet each computes the required actions for control based on either ongoing motor
signals or computational models that predict in advance states of the body and the environment.
Despite many decades of this research, the need for considering feedback did not go away. For
example, Optimal Control Theory was converted to a ‘hybrid’ - Optimal Feedback Control Theory –
to improve its match with observed data (Todorov & Jordan, 2002). Indeed, there are examples
across wide domains where models utilizing feedforward control rely critically on feedback
processing for accurate performance (Perkell, in press; Saunders & Vijayakumar, 2011).
The limitations of recent hybrid models are increasingly well documented, including their inability to
learn their control parameters, to engage in hierarchical control, and to control cognition
(Diedrichsen et al., 2010). Consequently, there have been recent calls for a more unified,
parsimonious theory of motor control that is directly testable (Arjemian & Hogan, 2010).
In this article, we heed this call and review evidence for a theory of feedback control that has the
capacity to restore confidence in the field: Perceptual Control Theory (PCT; Powers, Clark &
McFarland, 1960a,1960b; Powers, 1973, 2008). We make explicit four components of PCT
subsumed within its architecture that ‘feedforward’ their effects in parallel over four different time
scales (leaky integration of efferent signals, hierarchies, reorganization learning, and selfgenerated
feedback).
The article will initially critique five reasons commonly offered for the apparent advantages of
feedforward control within current „hybrid‟ models:
A. Signal delays make feedback control ineffective. This view seems inconsistent with evidence of
the performance advantages of sensory feedback during early stages of control, and for fast
movements (Boer et al., 2011; Tunik et al., 2009). Moreover, when disturbances are not
predictable, a model based on feedback processing can merely reduce its performance to the
same as that of a feedforward system. A feedback model with nerve signal delays performs
effectively over a range of simulated frequencies (Powers, 2008).
B. Sensory feedback is often unavailable. Under conditions of limited sensory feedback (e.g.
blocked vision), the brain seems capable of sensory substitution, whereby alternative streams
of sensory feedback, such as proprioception, are used (Merabet & Pascual-Leone, 2010).
C. Feedforward motor signals improve control. There are examples of feedforward motor signals,
such as the vestibulo-ocular reflex. Yet, these require stabilization through feedback processing
(Montfoort et al., 2008). The predicted sensory consequences of one’s own actions do,
however, appear to be used across a range of tasks; thus, a feedback theory needs to account
for this.
D. Control parameters for feedback systems need to be learned. Learning enhances control, yet
evidence shows that sensory feedback is typically required for this learning (Perkell, in press).
E. Internal models can predict the correct motor signal. The kinematic properties of the body and
environment are associated with specific brain regions (Grafton et al., 2008). Yet, internal
models can be complex and inaccurate (e.g. Cloete & Wallis, 2009). Thus, a parsimonious
account of internal model generation constantly updated by sensory feedback is desirable.
At this point within the article, accessible pop-out boxes will be used to explain PCT and the
evidence for the accuracy of models of behavioral tasks (Marken, 2009) and its explanation for
observed behavior (e.g. Pellis & Bell, 2011). In summary, the reference values for a feedback unit
in PCT are set by the efferent signals of a hierarchically superordinate feedback unit that operates
over a longer timescale of perceptual input. The whole system „controls perception‟, within motor
and environmental constraints. The lowest system of this hierarchy is the tendon reflex. An afferent
signal representing muscle tension is compared to an efferent reference signal entering the spinal
motor neuron, with the difference – error – being amplified by the muscle and converted to an
output force to maintain a regulated tension (Powers, 1973).
Importantly, there are four original features of PCT that could be conceptualized as ‘feed-forward’
and subserve its feedback architecture (Powers et al., 1960; Powers, 1973).
A pop-out box will
provide equations of these components and their location in a diagram. They are as follows:
(1) Leaky integration: efferent signals are integrated over periods of several milliseconds.
Therefore, the strength of a response to a disturbance in a constant direction will increase over
very brief periods. This counters signal delays (A) described above.
(2) Hierarchies: higher-level systems perceive changes that unfold over fractions of a second or
greater, and their efferent signals set the reference values for lower-level systems; these are
equivalent to the expected sensory consequences of action described above (C). For example, a
higher-level system for the desired angular position of a joint sets the desired velocity of the joint
for the system below. A mid-level system controls this variable via the efferent signals that it, in
turn, sends down to regulate acceleration via muscular forces at a lower level. Hierarchical PCT
models accurately match observed data (Marken, 1986; Powers, 2008). Future research could test
if a PCT model can utilize this information for a range of published tasks.
(3) Reorganization learning: during repeated or ongoing situations, the control parameters are
optimized through a trial-and-error learning process known as „reorganization‟, such that their
values are fed forward for improved performance. This addresses issue (D) described above.
Alternative streams of afferent input can develop through reorganization, addressing point (B).
Powers (2008) constructed and tested a computer model of 14-joint arm movement that learned its
optimal control parameters through this algorithm. Future research could test further models.
(4) Self-generated feedback: In PCT, when sensory feedback is unavailable and sensory
substitution has limited effectiveness, and/or when action is prevented, the control system can
enter ‘imagination mode’ (Powers, 1973). In this state, the reference values can be short-circuited
internally to the organism, so that approximate states of the self and world can be modeled ‘as if’
they are occurring in the environment. The motor plan that is modeled as most successful can then
be fed forward when the opportunity to control the environment is next available, subject to online
alterations through feedback control. This addresses point (E) above. Convergent evidence is
consistent with this process. Hierarchically organized closed loops of biafferent neural signals are
integral to the brain (Strick et al., 2011). Sensory deprivation in humans does indeed result in the
formation of internally generated perceptions such as spontaneous visual imagery (Mason &
Brady, 2009). Sensory deprivation in simulated robots leads to the emergence of spontaneous
internal generation of perception via a closed loop process (Gigliotta et al., 2010), and a study of
robotic systems utilizing sensory feedback models based on PCT outperformed competing models
by up to 95% (Rabinovich & Jennings, 2010).
A summary of several further advantages of adopting a PCT model of control will follow in the
article. First, a sense of control is recognised as an intrinsic need (Leotti et al., 2010), and
therefore it is appropriate that PCT places present-moment control, rather than prediction per se,
at its core. Second, cognitive control can be embodied within a PCT model (Mansell, 2011). Third,
the parsimony of PCT has made it highly adaptable to a variety of disciplines, e.g. mental health
(Carey, 2011) and organisational psychology (Vancouver & Scherbaum, 2008). Finally, the
limitations of PCT and directions for future research will be described.
Recent References
Ajemian, R., & Hogan, N. (2010). Experimenting with theoretical motor neuroscience. Journal of Motor
Behavior, 42, 333-342.
Carey, T. A. (2011). Exposure and reorganization: The what and how of effective psychotherapy. Clinical
Psychology Review, 31, 236-248.
Cloete, S. R., & Wallis, G. (2009). Limitations of feedforward control in multiple-phase steering movements.
Experimental Brain Research, 195, 481-487.
Diedrichsen, J., Shadmehr, R., & Ivry, R. B. (2010). The coordination of movement: optimal feedback control
and beyond. Trends in Cognitive Sciences, 14, 31-39.
Gigliotta, O., Pezzulo, G., & Nolfi, S. (2010). Emergences of an internal model in evolving robots subjected to
sensory deprivation. Lecture Notes in Computer Science, 6226, 575-586.
Grafton, S. T., Schmidt, P., Van Horn, J., & Diedrichsen, J. (2008). Neural substrates of visuomotor learning
based on improved feedback control and prediction. Neuroimage, 39, 1383-1395.
Leotti, L. A., Iyengar, S. S., & Ochsner, K. N. (2010). Born to choose: the origins and value of the need for
control. Trends in Cognitive Sciences, 14, 457-463.
Mansell, W. (2011). Control of perception should be operationalised as a fundamental property of the
nervous system. Topics in Cognitive Science, 3, 257-261.
Merabet, L. B., & Pascual-Leone, A. (2010). Neural reorganization following sensory loss: The opportunity of
change. Nature Reviews Neuroscience, 11, 44-52.
Marken, R. S. (2009) You say you had a revolution: Methodological foundations of closed-loop psychology.
Review of General Psychology, 13, 137-145.
Mason, O. J., & Brady, F. (2009). The psychotomimetic effects of short-term sensory deprivation. Journal of
Nervous and Mental Disease, 197, 783-785.
Montfoort, I., Van Der Geest, J. N., Slijper, H. P., De Zeeuw, C. I., & Frens, M. A. (2008). Adaptation of the
cervico- and vestibulo-ocular reflex in whiplash injury patients. Journal of Neurotrauma, 25, 687-693.
Tunik, E., Houk, J. C., & Grafton, S. T. (2009). Basal ganglia contribution to the initiation of corrective
submovements. Neuroimage, 47, 1757-1766.
Pellis, S., & Bell, H. (2011). Closing the circle between perceptions and behavior: A cybernetic view of
behavior and its consequences for studying motivation and development. Developmental Cognitive
Neuroscience, 1, 404-413.
Perkell, J. S. (in press). Movement goals and feedback and feedforward control mechanisms in speech
production. Journal of Neurolinguistics.
Saunders, I., & Vijayakumar, S. (2011). The role of feed-forward and feedback processes for closed-loop
prosthesis control. Journal of Neuroengineering and Rehabilitation, 8, 60.
Powers, W. T. (2008). Living Control Systems III: The Fact of Control. Benchmark Publications.
Rabinovich, Z., & Jennings, N. R. (2010). A hybrid controller based on the egocentric perceptual principle.
Robotics and Autonomous Systems, 58, 1039-1048.
Strick, P. L., Dum, R. P., Fiez, J. A. (2011). Cerebellum and Nonmotor Function. The Annual Review of
Neuroscience, 32, 413-434.
Vancouver, J. B. & Scherbaum, C. A. (2008). Do we self-regulate actions or perceptions? A test of two
computational models. Computational and Mathematical Organizational Theory, 14, 1-22.
Earlier References
Ashby, W. R. (1952). A Design for a Brain. Chapman & Hall.
Oosting, K. W. & Dickerson, S. L. (1987). Feed forward control for stabilisation. ASME
Powers, W. T. (1973). Behavior: The Control of Perception. Chicago, IL: Aldine.
Powers, W. T., Clark, R. K., & McFarland, R. L. (1960a). A general feedback theory of human behavior. Part
I. Perceptual and Motor Skills, 11, 71-88.
Powers, W. T., Clark, R. K., & McFarland, R. L. (1960). A general feedback theory of human behavior. Part
II. Perceptual and Motor Skills, 11, 309-323.
Todorov, E., & Jordan, M. I. (2002). Optimal feedback control as a theory of motor coordination. Nature
Neuroscience, 5, 1226-1235.
Wiener, N. (1948). Cybernetics: Or Control and Communication in the Animal and the Machine. MIT Press.
···
On Wed, Jan 29, 2014 at 4:35 AM, Richard H. Pfau richardpfau4153@aol.com wrote:
[From Richard Pfau (2014.01.28 23:36 EST)]
`Ref: [From Rick Marken (2014.01.28.1620)]
RM: Yes, how about writing an Ashby feed-forward model of the controlling a person does in a tracking task. Then let’s see how the model works in various versions of the task … Whaddaya think?
RP: Some feedback: Your suggestion for using a tracking task to test models of feed-forward and PCT feedback is highly biased in that the PCT model would surely perform better in such a tracking task.`
And a thought: In some cases, the phenomena called feedforward does seem to be taking place. (a) Bruce's example of the wrecking ball seems like one. (b) Another seems to be when you put on a coat when it's cold outside (i.e., before you are physically cold) to prevent yourself from becoming chilled when you go outside.
``
However, if as you indicated (2012.08.21.0915) and Bill Powers seemed to concur (2012.08.21.1217 MDT) that you "understood 'feedforward' to be basically equivalent to what we call a reference signal" so that [here's my interpretation now of what you seem to mean] (a) when someone shouts "Duck!" then reference signals (at the relationship and program levels?) are such that "such shouting (i.e., 'Duck!') indicates a relationship with danger" and should be quickly acted on, whereas (b)reference signals (again at the relationship and program levels?)for "when the weather is perceived to be cold, you should put on warm warm clothing when you go outside" results in the output of putting on a coat before going outside.
``
Are such interpretations more or less compatable with your thinking -- such that you feel that apparent feedforward phenomena are already covered by the PCT model?
``
With Regards,
Richard Pfau
–
Dr Warren Mansell
Reader in Psychology
Cognitive Behavioural Therapist & Chartered Clinical Psychologist
School of Psychological Sciences
Coupland I
University of Manchester
Oxford Road
Manchester M13 9PL
Email: warren.mansell@manchester.ac.uk
Tel: +44 (0) 161 275 8589
Website: http://www.psych-sci.manchester.ac.uk/staff/131406
See teamstrial.net for further information on our trial of CBT for Bipolar Disorders in NW England
The highly acclaimed therapy manual on A Transdiagnostic Approach to CBT using Method of Levels is available now.
Check www.pctweb.org for further information on Perceptual Control Theory