Motor Control I: The Brain as Controller
The previous chapters described the swing as a nonlinear mechanical system with drift, bounded actuation, feedback delay, and sensor noise. This chapter asks a narrower question: what kind of control architecture is plausible when the downswing lasts only a few hundred milliseconds?
The discussion treats the nervous system as a controller with feedforward planning, internal models, and delayed feedback. It does not claim that the brain solves an explicit engineering optimization problem on each swing. The useful claim is more limited: because mid-downswing correction is time constrained, skilled movement likely depends heavily on learned preparation, predictive control, and mechanical structure.
The Control Problem: What the Brain Must Solve
Recall from Chapter 6 and Chapter 6 that the golf swing is a control-affine system: \[ \dot{\bm{x}} = f(\bm{x}) + G(\bm{x})\bm{u}(t) \tag{1}\]
The state vector \(\bm{x}\) contains all joint angles and angular velocities. The drift term \(f(\bm{x})\) represents gravity and centrifugal effects—the physics does for free. The control term \(G(\bm{x})\bm{u}(t)\) represents what the muscles can accomplish. The control input \(\bm{u}(t)\) is the vector of muscle torques, subject to physical bounds: \(|\bm{u}| \leq \bm{u}_{\mathrm{max}}\).
The brain’s problem: compute \(\bm{u}(t)\) as a function of time to produce a desired motion that puts the ball at the target.
This is not a simple problem. It is a nonlinear optimal control problem in real time, subject to harsh constraints:
Neural delay: Motor commands must travel from the motor cortex to the muscles via the spinal cord. Typical conduction time: 30–50 milliseconds. If the downswing lasts 300 milliseconds, the motor command that initiates the acceleration phase must be generated 50 milliseconds before the transition.
Activation dynamics: Once a motor neuron fires, the muscle doesn’t instantly produce torque. Muscle force rises with a time constant of 50–100 milliseconds. The brain cannot produce torque changes faster than this.
Sensory feedback delay: Sensorimotor feedback is not one serial channel. Short-latency stretch responses begin at roughly 20–45 ms, long-latency responses at roughly 50–100 ms, and voluntary responses at greater than 100 ms in the upper limb; visually guided responses usually enter later and depend on the task and response being measured (Kurtzer 2014; Pruszynski and Scott 2012). These are exemplar onset bands, not universal golf-swing constants.
Noisy sensors: Proprioceptive acuity varies by joint. At the wrist, you can detect a 1–2 degree change in angle. At the shoulder, it’s 3–5 degrees (Proske and Gandevia 2012). Vision has high acuity but slow feedback. The nervous system must work with uncertain, delayed measurements.
Noisy actuators: Muscle force variability is 5–10% even when you’re trying to produce exactly the same torque twice. This is called signal-dependent noise and increases with force level.
Bounded control: Maximum torques are limited. In this illustrative range, the shoulder can produce roughly 100–200 N\(\cdot\)m of torque, the elbow 100–150 N\(\cdot\)m, the wrist 30–50 N\(\cdot\)m (Enoka 2002). The brain cannot exceed these.
In an illustrative 250–300 ms downswing, several parallel pathways may contribute:
- Short-latency stretch response: approximately 20–45 ms
- Long-latency response: approximately 50–100 ms
- Voluntary response: greater than 100 ms
- Muscle activation and force development: additional, muscle- and task-dependent dynamics
- Visually guided response: typically late relative to the downswing, with timing and mechanical effect dependent on stimulus, response definition, and swing phase
The onset bands must not be added into one “round-trip” total: the pathways are partly parallel and close through different circuitry (Kurtzer 2014; Pruszynski and Scott 2012). Timing alone also does not establish response authority—whether a response begins before impact is different from whether it can materially change club or body state. The defensible conclusion is narrower: late visually guided gross corrections are strongly constrained, while short- and long-latency feedback may still regulate muscle activity, limb impedance, or a local trajectory error. Which effects matter in golf remains a phase- and task-specific empirical question.
Feedforward and Feedback: Complementary Contributions
Feedforward and feedback are useful analytical components, but biological control can combine them rather than choosing one exclusively:
The controller measures the current state \(\bm{x}(t)\), compares it to the desired state \(\bm{x}_d(t)\), computes an error \(\bm{e}(t) = \bm{x}_d(t) - \bm{x}(t)\), and adjusts the control input based on this error: \[ \bm{u}(t) = \bm{K}\bm{e}(t) = \bm{K}[\bm{x}_d(t) - \bm{x}(t)] \] where \(\bm{K}\) is a gain matrix.
This is reactive: if things are going wrong, the controller corrects.
The controller uses a model of the system to compute the required control input in advance: \[ \bm{u}(t) = \bm{u}_{\mathrm{planned}}(t) \]
This component is prepared before the corresponding sensory consequence arrives. It can operate alongside feedback during execution.
The useful hypothesis is that learned preparation and feedforward commands carry substantial responsibility for the downswing, while feedback remains pathway- and phase-dependent.
For example, a new visual error presented at \(t=100\) ms may not produce a mechanically important correction before a \(t\approx300\) ms impact. That scenario bounds one late visually guided loop; it does not rule out faster proprioceptive pathways or responses already in progress. Long-latency responses are task-dependent and can incorporate limb mechanics and behavioral goals (Kurtzer 2014; Pruszynski and Scott 2012).
Unsupported extreme: A golfer continuously sees clubface error and fully replans the downswing in real time.
Bounded interpretation: A late visual perturbation is unlikely to support a complete replan before impact. Faster feedback can still modulate the ongoing action, and post-shot feedback supports learning. The response authority of each pathway depends on perturbation onset, swing phase, task, response measure, and mechanical outcome.
Short- and long-latency proprioceptive responses can begin within a downswing, but onset is not equivalent to a successful clubhead correction. Establishing their golf-specific role requires time-locked perturbations, muscle and motion measurements, and an outcome definition. Until golf-specific perturbation evidence is available, claims about elite golfers suppressing or relying on a pathway are hypotheses, not established facts.
This timing picture motivates careful study of pre-swing preparation, initial conditions, impedance, and learned commands without assuming that execution is open-loop or deterministic.
Internal Models: The Brain’s Simulator
If the golf swing is feedforward, how does the brain know what command to send? It has a model. In this illustrative model, the brain maintains an internal model—a neural simulation of the body’s dynamics.
A forward model predicts the sensory consequences of a motor command. Given the current state \(\bm{x}(t)\) and a motor command \(\bm{u}(t)\), the forward model predicts the next state: \[ \hat{\bm{x}}(t+\Delta t) = \hat{f}(\bm{x}(t)) + \hat{G}(\bm{x}(t))\bm{u}(t) \]
Mathematically, it IS solving the nonlinear dynamics equation. The brain becomes a physicist in practice, even if not in theory.
An inverse model computes the motor command needed to achieve a desired sensory outcome. Given a desired state change \(\bm{x}_d(t)\), the inverse model computes: \[ \bm{u}(t) = \hat{G}(\bm{x}(t))^{-1}[\dot{\bm{x}}_d(t) - \hat{f}(\bm{x}(t))] \]
This is the inverse dynamics problem: “What muscles do I fire to move the club from here to there?”
The forward and inverse models are intimately related. The brain learns them together. A forward model is easy to differentiate numerically to obtain an approximate inverse model: \(\bm{u} \approx \hat{G}^{-1}(\bm{x}_d - \hat{f})\).
The Cerebellum as the Seat of Internal Models
Where in the brain do these internal models live? The leading candidate is the cerebellum, a structure at the base of the brain.
The cerebellum is remarkable for several reasons:
Size: The cerebellum is approximately 10% of the brain’s volume, but contains an estimated 80% of all neurons—about 69 billion of the brain’s 86 billion, the overwhelming majority of them granule cells (Azevedo et al. 2009). That makes it one of the most neuron-dense structures in the brain. It also contains roughly 15 million Purkinje cells, the large output neurons that project to the deep cerebellar nuclei (Eccles 1973).
Computational density: This makes the cerebellum among the most densely packed neural tissue in the brain—a parallel processor of extraordinary capacity.
Stereotyped circuit: The cerebellar circuit is highly organized and unusually well characterized compared with many cortical circuits, although it is not fully understood.
Learning rule: The cerebellum uses a clear, identified learning mechanism: long-term depression (LTD) of synapses. When a Purkinje cell receives simultaneous input from a climbing fiber (error signal) and parallel fibers (motor plan), the parallel fiber–Purkinje synapses are weakened. This is an error-correction learning rule.
Timing: The cerebellum operates fast, with learning that can occur within a single trial for simple adaptation tasks.
Cerebellar damage can produce severe loss of motor coordination—a condition called ataxia. Patients with cerebellar ataxia often struggle to perform smooth, coordinated movements, including reaching tasks and skilled striking actions. This is consistent with impaired prediction, timing, and error-correction mechanisms, though the clinical picture varies by lesion and task.
The cerebellar learning rule is thought to implement the following algorithm: \[ \Delta w_{ij} = -\eta \cdot e(t) \cdot x_i(t) \cdot y_j(t) \tag{2}\]
where \(w_{ij}\) is the synaptic weight between parallel fiber \(i\) and Purkinje cell \(j\), \(e(t)\) is the climbing fiber error signal, \(x_i(t)\) is the firing rate of parallel fiber \(i\), and \(y_j(t)\) is the firing rate of the Purkinje cell. The negative sign indicates that errors drive depression of relevant synapses. Over time, this makes the forward model more accurate.
This is a simplified form of the Marr-Albus-Ito hypothesis. More precisely, the climbing fiber carries an error signal, and parallel fiber synapses that are co-active with climbing fiber input undergo long-term depression (LTD). The learning rule above captures the essential structure: weights decrease when error and pre-synaptic activity coincide (Wolpert et al. 1995; Flanagan et al. 2003; Kawato 1999).
Why Internal Models Matter
Without prediction, the brain would have limited ability to plan a fast golf swing. Here’s why:
Suppose you did not use a predictive model. You swing the club and see, after a visual delay, that the clubface was open. Any correction based only on that delayed information refers to a previous body and club state. Because the dynamics are nonlinear, the effect of a torque at position \(\bm{x}_1\) differs from the effect of the same torque at position \(\bm{x}_2\). Delayed feedback can still support learning and early corrections, but it is poorly suited to precise late-downswing adjustment.
With an internal model, the nervous system can predict likely sensory consequences before they occur. If the planned motion predicts an open clubface, the plan can be adjusted before or early in the swing.
The forward model enables predictive control. The inverse model enables planning. Together, they enable the feedforward computation of motor commands that reliably produce desired outcomes.
The Predictive Processing Framework
Predictive-processing theories model the brain as continuously generating and updating predictions about sensory input. The free energy principle and related predictive-processing frameworks are influential accounts, but they are not the only active theories of motor control (Friston 2010; Clark 2013).
The brain doesn’t passively receive sensory information. Instead, it constantly generates predictions about what it expects to sense. Perception works by comparing predictions with actual sensations. When they match, the brain’s model of the world is good. When they don’t match, a prediction error occurs, and the brain updates its model.
For the golf swing, this framework explains several phenomena:
Pre-swing planning: Before you swing, your brain generates a predicted trajectory. This is your forward model running forward: “If I fire these muscles at these times, the club will follow this path, and the ball will land there.” This is an active inference—you’re inferring which motor command will make the world match your prediction (the target).
During-swing monitoring: As you swing, your brain continuously compares predicted sensory signals with actual signals. The proprioceptive feedback is compared against the forward model’s predictions. If they match, all is well. If they don’t, there’s a prediction error.
Post-swing learning: After the swing, a large prediction error is available: the ball lands at location \(\bm{x}_{\mathrm{actual}}\) instead of the predicted location \(\bm{x}_{\mathrm{predicted}}\). This error drives learning. The brain updates its internal model to reduce the error on the next swing.
Limited time for some corrections: A 300 ms downswing constrains late voluntary and visually guided responses, while faster parallel pathways can still contribute. Prior learning is important, but the number and effect of within-swing responses cannot be inferred by dividing duration by one nominal delay.
This framework has profound implications. It says that action is not fundamentally different from perception. Both work through prediction and error correction. The goal of a motor action is to make the actual sensory consequences match your predicted sensory consequences. For the golf swing, the predicted sensory consequence is the feeling of a well-struck shot—the proprioceptive pattern that you’ve learned corresponds to good contact.
Ideomotor Theory: Think the Result, Not the Process
William James, in his 1890 Principles of Psychology, proposed a radical idea: actions are represented by their effects, not by their motor commands. When you think about throwing a ball, you think about the trajectory of the ball, not the sequence of muscle contractions. Your motor system then computes the commands needed to produce that trajectory.
Modern ideomotor theory (Prinz 1997; Hommel et al. 2001) has formalized this insight:
An action is coded as a set of desired sensory consequences (the effect). When the action goal is activated, the motor system automatically retrieves the motor commands that, in the past, produced those consequences. The action is executed by commands designed to minimize discrepancy between the predicted sensory consequences and the desired consequences.
This has a striking implication for golf coaching: the coach should focus the golfer’s attention on the desired outcome, not on the mechanics of movement.
Consider two coaching approaches:
Mechanics-focused: “Keep your head still. Shift your weight to your front side. Lag the club. Extend through the ball.”
Outcome-focused: “Feel the swing shape. See the ball flight. Feel the contact. Swing to the target.”
According to ideomotor theory, the second approach is more effective. Here’s why: when you think about mechanics, you’re thinking about motor commands (“fire the pectoral muscle”, “lag the club”). This bypasses the action effect pathway. Your motor system is confused—it’s trying to both retrieve learned commands and generate new commands based on mechanical rules.
When you think about the outcome (the ball flight, the feeling of good contact), you activate the action effect. Your motor system automatically retrieves the learned motor program that produces that effect. No confusion. No conflict.
This explains why great coaches talk about feel and target, not mechanics.
The ideomotor principle also connects to the drift field. In the ZTCF family framework, the brain doesn’t need to control every detail of the trajectory. Instead, it specifies the initial conditions and lets physics (the drift field) carry the system to the outcome. In ideomotor terms, the brain specifies the desired sensory consequence (impact with the ball, ball trajectory), and the motor system retrieves the commands that, given physics, produce that consequence (Prinz 1997; Hommel et al. 2001; Wolfensteller and Ruge 2011).
The Nervous System Architecture: Hardware Specifications
To understand how the brain controls movement, we need to understand its hardware architecture. The nervous system is not a single, monolithic controller. It’s a hierarchy of controllers, each operating at a different timescale and with different capabilities.
The Motor Cortex
The primary motor cortex (M1) is located in the frontal lobe, just in front of the central sulcus. It contains roughly 100,000 corticospinal neurons per hemisphere, each sending axons down through the spinal cord to synapse onto motor neurons.
The major pathway for voluntary motor control. Fibers originate in the motor cortex, travel through the brain and brainstem, cross the midline (decussate) in the medulla, and descend the spinal cord. Each M1 neuron controls the muscles on the opposite side of the body.
Signal transmission in these fibers is fast—roughly 120 m/s in myelinated fibers. This means a motor command takes about 30 ms to reach the hand muscles, and about 50 ms to reach the foot muscles (Kandel et al. 2013).
An important principle:
The motor cortex doesn’t work one neuron per muscle. Instead, it uses population coding. Thousands of neurons, each with slightly different directional preferences, are active simultaneously. The vector sum of their activity encodes the direction and magnitude of the desired motion. This population code is robust: loss of individual neurons has little effect because the information is distributed.
The Spinal Cord: Local Control Centers
The spinal cord is not just a cable carrying commands down and feedback up. It contains local circuits that can coordinate movement independently of the brain.
A neural circuit that produces rhythmic output without rhythmic input. CPGs in the spinal cord generate patterns like walking, swimming, or scratching. They can operate autonomously, though they’re normally modulated by brain signals.
A rapid response to muscle stretch. When a muscle is stretched, spindle receptors send a signal directly to motor neurons in the spinal cord, which fire and contract the muscle. This entire loop happens in about 30 milliseconds, without brain involvement. The brain doesn’t know about it until after it happens.
For the golf swing, short-latency spinal pathways may contribute local stabilizing responses to a perturbation. Longer-latency supraspinal and visually guided contributions would enter later. Whether any pathway materially changes the club state requires a defined perturbation, phase, and measured outcome.
The Basal Ganglia: Action Selection
The basal ganglia are a set of nuclei deep in the brain. Their primary function is action selection—deciding which motor program to execute.
A learned sequence of muscle activations that produces a stereotyped action (like a golf swing). Once selected and initiated, the program unfolds with minimal brain intervention.
The basal ganglia use a neurochemical signal called dopamine to encode reward prediction errors. When you execute a motor program and it produces a better-than-expected outcome (a good shot), dopamine neurons fire. This signal reinforces the basal ganglia circuits that led to that action. Over time, actions that produce rewards become more likely.
The pre-shot routine involves several motor programs:
- The waggle (small oscillations of the club)
- The step(s) to position
- The alignment checks
- The final address
Each of these is a learned motor program. The basal ganglia select which routine to execute and in what order. Once the routine is selected, it unfolds automatically.
When you hit a great shot, dopamine neurons fire (reward). The basal ganglia strengthen the circuits that selected that particular pre-shot routine. Over time, the routine becomes more automatic.
When you hit a poor shot, dopamine neurons are quiet (no reward). The circuits that selected that routine weaken. You’re less likely to use that routine next time.
This is reinforcement learning implemented in neural circuits.
Parkinson’s disease, which damages dopamine neurons, produces profound difficulty in action selection. Patients know what they want to do, but they cannot initiate movement. They can stand frozen, unable to take a step. This illustrates how critical the basal ganglia are for movement initiation.
Motor Unit Recruitment: Henneman’s Size Principle
A motor unit is a motor neuron and all the muscle fibers it innervates. A single motor neuron controls 10–1000 muscle fibers, depending on the muscle. The smaller the motor neuron, the fewer fibers it controls.
Motor neurons are recruited in order of size: small (slow, weak) units first, large (fast, strong) units last (Henneman et al. 1965). This ordering is automatic and doesn’t require brain control. It’s determined by the electrophysiology of the motor neuron pool.
This principle has important consequences:
You can’t produce maximum force instantly. To achieve high forces, you must recruit the large motor units. But you can’t recruit them without first recruiting all the small units. This takes time—typically 100–500 milliseconds.
At low effort, you have precision control. With only small units active, forces are fine-graded and controllable. This is why you can write with a pen or play piano (low forces, high precision).
High effort can constrain force grading. Recruiting larger motor units changes the available force increments and signal-dependent variability. The effect on movement precision depends on task and coordination; it is not a universal proof that maximum-effort movement is ballistic.
The golf swing transitions from low force to high force. During setup and the backswing, forces are moderate and the system is highly controllable. As the transition approaches, forces increase. By the downswing, the system is in a high-force regime where ballistic execution is the predominant mode.
Bandwidth Limitations: The Brain’s Processing Speed
A fundamental constraint on any controller is its bandwidth—the rate at which it can process information and generate corrections.
The maximum frequency at which a controller can detect errors and generate corrective responses. For the brain, the limiting factor is not neural transmission speed (which is fast—30–120 m/s) but the time taken to process information and compute responses.
The conscious brain updates motor plans at roughly 4–10 Hz (one update every 100–250 ms). Visual flicker fusion occurs at \(\sim60\) Hz, but conscious decision-making and motor replanning are far slower. This mismatch is critical: sensory data arrives faster than the brain can act on it.
A light flickering at 60 Hz appears steady to your eye, even though it’s actually off half the time. This is because 60 Hz exceeds your conscious bandwidth. Each on–off cycle is too fast to perceive as distinct.
For the golf swing, this bandwidth limitation is severe:
A 300 millisecond downswing at 10 Hz bandwidth means the brain can process roughly 3 distinct time windows during the swing:
- \(t=0\)–100 ms: transition phase (beginning of acceleration)
- \(t=100\)–200 ms: acceleration phase (maximum force)
- \(t=200\)–300 ms: deceleration phase (approaching impact)
This three-window sketch is a coarse pedagogical partition, not a count of feedback cycles. Different feedback pathways overlap these windows and have different response authority.
It is tempting to blame a bandwidth mismatch — a controller too slow for a fast plant — but the numbers do not support that reading. A driver shaft’s first bending mode is about 3–5 Hz, and the gross swing dynamics are slower still: the whole downswing is a single acceleration–deceleration cycle lasting roughly a quarter of a second. The plant is therefore slower than the 10–40 Hz at which the nervous system can modulate its output, not faster.
Transport delay is an important constraint, but it must be interpreted by pathway. A short-latency response can begin at roughly 20–45 ms, a long-latency response at 50–100 ms, and a voluntary response at greater than 100 ms (Kurtzer 2014; Pruszynski and Scott 2012). Late visually guided gross corrections may have little response authority before impact, while earlier proprioceptive feedback may modulate impedance or local trajectory. Timing alone does not determine the mechanical outcome.
This is the classic problem of a control loop whose delay is comparable to the duration of the task itself.
One model-based response is to exploit drift and mechanical structure. When a motion is partly drift-dominated, a slow controller can perform better by preparing initial conditions, stiffness, and feedforward commands rather than relying on moment-by-moment correction. This does not mean initial conditions are the only thing that matters, but it explains why preparation and learned timing are central (Todorov 2005; Shadmehr and Wise 2005).
The Brain’s Processing Power: Energy and Computation
How much “horsepower” does the brain have for controlling movement?
Energy consumption: The brain uses roughly 20 watts of power at rest, about 20% of the body’s total metabolic rate. This seems like a lot, but it’s actually quite efficient for the amount of computation being performed.
Computational capacity: Estimates of the brain’s computational capacity vary widely depending on what counts as an ‘operation.’ If each of the brain’s \(\sim 10^{14}\) synapses fires at \(\sim10\) Hz, the brain performs roughly \(10^{15}\) synaptic operations per second (Wolpert et al. 2011). Under more liberal definitions (including subthreshold integration), estimates reach \(10^{16}\)–\(10^{18}\). These numbers are uncertain, but the order of magnitude is instructive.
Per-neuron computation: With \(\sim86\) billion neurons (Azevedo et al. 2009), this is roughly \(10^4\)–\(10^5\) synaptic operations per neuron per second. Direct comparison with digital processors is misleading: neurons are analog, parallel, and asynchronous, while CPUs/GPUs perform discrete, clocked operations. The brain compensates for slower individual operations with massive parallelism.
The brain need not solve the explicit engineering optimization written here at every instant. Learned policies, state estimation, prediction, and task-dependent feedback can all contribute online.
A robotic arm might use real-time trajectory optimization: at each time step (every millisecond), solve a local optimal control problem to compute the next motor command. This requires substantial computation per time step.
Biological control can learn over repeated trials and reuse prepared policies during execution. That does not imply simple playback: prediction and task-dependent feedback can modify the ongoing command.
Robot and human adaptation rates depend on controller, task, perturbation, and learning history. This comparison is conceptual; it is not a measured energy or adaptation-time benchmark.
Practice can shape prepared commands and feedback policies. During a swing, their relative contributions remain an experimental question rather than a strict offline/online divide.
Learned preparation and feedforward commands are likely important in a short downswing, but feedback remains active through parallel pathways. The mechanical response authority of each pathway depends on latency, phase, task, and outcome.
The brain controls movement using internal models—forward models that predict sensory consequences, and inverse models that compute motor commands. The cerebellum is the likely neural substrate for these models.
The brain operates as a predictive processing system. It constantly generates predictions and compares them to actual sensory input. Prediction errors drive learning.
Actions are represented by their effects (ideomotor theory), not by motor commands. The brain specifies a desired outcome, and the motor system retrieves the commands that produce that outcome.
The nervous system is a hierarchy: the motor cortex (high-level planning), the spinal cord (local control), and the basal ganglia (action selection) all contribute.
Delay and plant dynamics jointly constrain control. Late visually guided gross corrections may have little time before impact; faster proprioceptive pathways may still modulate the ongoing action. Golf-specific perturbation evidence is needed to quantify those contributions.
Practice shapes prepared commands, state estimation, and feedback policies that can all contribute during performance; this is not a strict offline-learning/online-playback split.
Chapter Exercises
The nervous system modulates output at 10–40 Hz, while the shaft’s first bending mode is only 3–5 Hz — so the controller is faster than the plant. Why, then, can the golfer not correct the swing in flight? Frame your answer in terms of loop delay against movement duration rather than bandwidth, and state which reflex loops could close within a 250 ms downswing and what they could actually accomplish.
A professional golfer produces clubhead speeds with a coefficient of variation of 1–3% (Jorgensen 1994). Given the neural noise (5–10% muscle force variability) and sensory noise ( 1–5 degree proprioceptive acuity), explain how this high consistency is possible. Hint: what does the drift field contribute?
Suppose you wanted to teach a robot to swing a golf club using online optimal control (computing the motor command at each millisecond). What would be the minimum computation time required? Assume the downswing lasts 300 ms and the robot must solve a 20-dimensional optimal control problem at each millisecond.
In Section 1.5.4, we discussed Henneman’s size principle: motor units are recruited in order of size. Why is this principle important for control? What would happen if large motor units were recruited first?
Compare a short-latency proprioceptive response, a long-latency response, and a late visually guided response during a 300 ms downswing. What additional measurement is needed to determine whether each response has useful mechanical authority?
According to ideomotor theory, what should a golf coach emphasize: mechanical details of the swing, or the desired outcome and feel? Why?
The cerebellum contains 70% of all brain neurons but is only 10% of brain volume. What does this suggest about the cerebellum’s computational strategy?
A golfer with cerebellar damage cannot hit a golf ball. They can move their arms, and they can understand the task. What is missing? How does this relate to the concepts of forward and inverse models?
Sketch a block diagram of the golf swing control system. Include the motor cortex, motor command, muscles, dynamics, sensory feedback, and feedback delays. Indicate which components operate on which timescale.
Compare feedforward and feedback components in a golf swing. Which pathways could respond within a 250–300 ms downswing, and what experiment would distinguish response onset from useful mechanical correction?