Passive Distributed Control: Frans Bosch Framework
Overview
Frans Bosch’s Strength Training and Coordination (2015) presents a radically different view of athletic movement: the body does not issue commands to muscles; it creates conditions for self-organization. The central nervous system acts more as a constraint relaxer than a motor commander.
This framework aligns remarkably well with the control-affine drift-control decomposition of Volume I: the “drift” is not a disturbance to be rejected — it is the self-organizing passive dynamics that expert athletes learn to exploit.
“The movement system acts as a distributed, self-organizing network. Efficient movement emerges from the coordination of passive elements — tendons, fascial structures, gravitational forces, inertial effects — rather than from active muscular force production.”
In mathematical terms: \(\dot{\mathbf{x}} = f(\mathbf{x}) + G(\mathbf{x})\mathbf{u}\), and expert performance means \(\|\mathbf{u}\| \to 0\) (control effort approaches zero as passive dynamics \(f(\mathbf{x})\) do the work).
Self-Organization in Complex Movement
Attractors as Emergent Properties
A motor attractor is a recurring movement pattern that emerges without explicit programming. Dynamical systems theory (Kelso, 1995) shows that biological movement patterns are attractors in the space of coordination variables — states to which the system naturally converges.
In the control-affine model, the golf swing downswing exhibits attractor-like behavior: once the backswing position and initial velocity are established, gravitational loading, elastic energy in stretched musculotendinous units, and inertial coupling from the proximal-to-distal sequential chain contribute substantially to the downswing dynamics. EMG studies indicate that muscular activation is present throughout the downswing, but the passive contributions may dominate the late-downswing acceleration profile—a hypothesis testable through inverse dynamics analysis.
Connection to Vol I: The attractor manifold corresponds to the stable portion of the contraction manifold from Chapter 4. Points near the attractor converge exponentially with contraction rate \(\lambda\).
Differential Learning (Schöllhorn, 2000)
Conventional coaching repeats the “correct” movement to build muscle memory. Differential learning (DL) deliberately introduces random perturbations to every repetition — no two swings are the same.
Paradoxically, DL produces superior learning outcomes (Schöllhorn et al., 2006) because:
- Prevents premature attractor formation: Constant perturbation keeps the learner exploring the attractor basin rather than converging to a local minimum
- Trains robustness: The motor system learns to return from many nearby states, not just one
- Exploits implicit learning: Variability exploration allows the cerebellum to discover the efficient passive dynamics without explicit instruction
Mathematical interpretation: DL is a noise-injection strategy that prevents the covariance matrix of the state distribution from collapsing to zero. Maintains a minimum exploration radius \(\sigma > \sigma_{\min}\).
Bernstein’s Three Stages
Nikolai Bernstein (1967) described skill acquisition in three stages, now understood through the passive dynamics lens:
Stage 1: Freezing Degrees of Freedom
Beginners reduce the effective DoF by stiffening joints. A novice golfer grips the club tightly and restricts wrist movement. This makes the system more predictable (reducing model uncertainty) at the cost of efficiency.
Control-theoretic view: High feedback gain \(K \to \infty\) effectively “freezes” a DoF by suppressing any deviation from the reference. This is expensive in control effort but reduces state uncertainty.
Stage 2: Freeing Degrees of Freedom
As the forward model improves, the learner progressively releases joints. The wrist “cocks” and uncocks; the hips lead the shoulders. Additional DoF are exploited one by one.
Control-theoretic view: Reducing gain \(K\) on progressively unfrozen DoF. The effective control bandwidth decreases as the forward model takes over prediction.
Stage 3: Exploiting Degrees of Freedom — Passive Dynamics
The expert exploits elastic rebound, gravitational torques, and inertial coupling effects. The late wrist release in the golf swing appears to be driven primarily by passive dynamics — proximal-segment deceleration transferring angular momentum distally — rather than active muscular effort at the wrist, though the relative contributions remain an active area of research.
Control-theoretic view: \(\mathbf{u} \to 0\). The drift term \(f(\mathbf{x})\) does all the work. The drift-control ratio \(\rho \gg 1\).
Passive Walking and McGeer’s Insight
Tad McGeer (1990) built a passive walking machine with no motors, no control, and no sensors. Powered only by gravity on a gentle slope, it walks stably for hundreds of steps. This demonstrates that walking is a passive attractor of bipedal mechanical systems — the controller’s job is only to maintain conditions for this attractor.
This insight shaped a lineage of efficient walking robots. The most direct descendants are the actuated passive-dynamic walkers (Collins, Ruina, Tedrake & Wisse, 2005; Cornell Ranger), which add only small amounts of power to a fundamentally passive gait. More heavily actuated platforms (Boston Dynamics Atlas, MIT Cheetah) do not “walk passively” — Atlas in particular is hydraulically actuated and relies on model-predictive control — but they still benefit from designing the mechanics so that natural dynamics and series-elastic compliance do much of the work, leaving control to handle disturbance rejection and foot placement.
Key result: The passive walking gait is a limit cycle in state space. Its existence and stability are properties of the mechanical system, not the controller. The Poincaré return map has a fixed point; stability is determined by the eigenvalues of the linearized return map — exactly the analysis of Chapter 4, Volume I.
Elastic Energy Storage: The Catapult Effect
Muscles and tendons form a muscle-tendon unit where: - The muscle provides the force (contractile element) - The tendon stores and releases elastic energy (spring element)
In powerful ballistic movements (throwing, jumping, swinging), the optimal strategy is:
- Load the spring: Stretch the tendon under muscular force (eccentric contraction)
- Freeze the muscle: Momentarily hold the stretched position
- Release: Tendon recoil propels the distal segment at speeds exceeding what muscular contraction alone could produce
The Achilles tendon can return on the order of 35% of the mechanical energy required for each running step (Ker et al., 1987), recoiling during push-off rather than that energy being supplied freshly by muscle. (The exact fraction depends on speed and on how the energy budget is defined, so treat 35% as a representative figure, not a universal constant.) By analogy, expert golfers are thought to store and return elastic energy in the Achilles, patellar, and wrist-extensor tendons across the backswing-to-downswing transition.
Mathematical model: The muscle-tendon unit as a series spring-mass-damper:
\[m\ddot{x}_m + b\dot{x}_m + k_m(x_m - x_t) = F_{\text{active}},\] \[k_t(x_t - x_{\text{ext}}) = k_m(x_m - x_t),\]
where \(x_m\) is muscle element position, \(x_t\) is tendon attachment point, \(k_t\) is tendon stiffness, and \(F_{\text{active}}\) is active muscle force. The energy stored in the tendon is \(E_t = \frac{1}{2}k_t \delta_t^2\) where \(\delta_t = x_t - x_{\text{ext}}\).
Fascial Tensegrity
The fascia — the connective tissue network enveloping all muscles — transmits force across multiple joints and segments simultaneously. Bosch emphasizes that force transmission is not purely through muscles but through the fascial net:
- A hip extension movement is proposed to transmit tension through the plantar fascia up to the thoracolumbar fascia, along what Myers (2001) calls “Anatomy Trains”
- On this view, the distributed tension network enables force summing across multiple segments
It is worth flagging that the “Anatomy Trains” myofascial-meridian model is a popular and clinically influential framework rather than a rigorously validated one; the degree to which appreciable force is transmitted along these fascial lines (versus through bones and joints) remains debated in the biomechanics literature. We use it here as a motivating analogy for multi-segment coupling, not as established fact.
Control implication: The effective “control input” to a distal segment is not just the locally acting muscles but the integral of forces transmitted through the fascial chain. The \(G(\mathbf{x})\) matrix in the control-affine model must account for these multi-segment coupling effects.
Representing fascial transmission as off-diagonal entries in \(G(\mathbf{x})\) is a speculative extension, not a settled result. Quantitative calibration of those entries from imaging or dissection data is an open problem. Most of the passivity analysis in the rest of this article treats fascia effects conservatively — as contributions to the mass and damping matrices \(M(\mathbf{q})\) and \(D(\mathbf{q}, \dot{\mathbf{q}})\) that enter the drift term \(f(\mathbf{x})\), rather than as modifications to the input map \(G(\mathbf{x})\). Readers adopting the off-diagonal-\(g\) view should regard it as a research hypothesis.
Cross-Volume Integration
The “Vol I”, “Vol II”, etc. references in this table point to chapters across the four volumes of The Geometry of Motion published on AffineDrift (see Books Index). The table is included to show the structural connections between the AffineDrift articles and the textbook series.
| Volume | Bosch Framework Connection |
|---|---|
| Vol 0, Ch 6-7 | Recursive dynamics algorithms handle multi-segment force transmission |
| Vol I, Ch 3 | Superposition principle: drift contribution from elastic storage |
| Vol I, Ch 4 | Contraction metrics for passive walking limit cycles |
| Vol I, Ch 7 | Counterfactual: what happens without control = pure passive dynamics |
| Vol II, Ch 5 | Spring-damper elements in multibody models |
| Vol III, Ch 3 | Optimal timing of elastic energy release (bang-bang control) |
Python: Passive Dynamics Simulation
import numpy as np
from scipy.integrate import solve_ivp
import matplotlib.pyplot as plt
def muscle_tendon_unit(t, state, k_t, k_m, b_m, m, F_active_func):
"""
Muscle-tendon unit dynamics.
state = [x_m, v_m, x_t] (muscle position, velocity; tendon attachment)
"""
x_m, v_m, x_t = state
F_active = F_active_func(t)
# Tendon force (spring)
F_tendon = k_t * x_t # x_ext = 0 equilibrium
# Muscle force balance (tendon pull = muscle spring + damper + active)
# k_m(x_m - x_t) = F_active - m*x_m_ddot - b_m*v_m
# F_tendon = k_m(x_m - x_t)
a_m = (F_active - b_m * v_m - k_m * (x_m - x_t)) / m
v_t = (k_m * (x_m - x_t) - k_t * x_t) / (b_m + 1e-3) # quasi-static tendon
return [v_m, a_m, v_t]
# Parameters: wrist-club tendon during downswing
k_t = 5000.0 # N/m tendon stiffness
k_m = 1000.0 # N/m muscle spring stiffness
b_m = 50.0 # Ns/m muscle damping
m_club = 0.45 # kg club head effective mass
# Active force profile: load during backswing, release at downswing start
def F_active(t):
if t < 0.2: # Backswing: eccentric loading
return 100.0
elif t < 0.25: # Freeze: isometric hold
return 80.0
else: # Release: passive recoil
return 0.0
state0 = [0.02, 0.0, 0.01] # Initial stretch
sol = solve_ivp(
lambda t, y: muscle_tendon_unit(t, y, k_t, k_m, b_m, m_club, F_active),
(0, 0.5), state0, max_step=0.001, dense_output=True
)
# Energy stored and released
E_tendon = 0.5 * k_t * sol.y[2]**2
E_muscle_spring = 0.5 * k_m * (sol.y[0] - sol.y[2])**2
E_kinetic = 0.5 * m_club * sol.y[1]**2
plt.figure(figsize=(10, 5))
plt.plot(sol.t, E_tendon * 1000, label="Tendon elastic energy (mJ)")
plt.plot(sol.t, E_kinetic * 1000, label="Kinetic energy (mJ)")
plt.axvline(0.25, color="r", linestyle="--", label="Release point")
plt.xlabel("Time (s)"); plt.ylabel("Energy (mJ)")
plt.title("Muscle-Tendon Unit: Elastic Storage and Release")
plt.legend(); plt.grid(True)Summary
| Bosch Concept | Mathematical Equivalent | Book Location |
|---|---|---|
| Passive attractor | Stable limit cycle; \(\lambda < 0\) | Vol I, Ch 4 |
| DoF freezing | High gain feedback \(K \to \infty\) | Vol I, Ch 5 |
| DoF exploitation | Drift dominance \(\rho \gg 1\) | Vol I, Ch 8 |
| Elastic storage | Spring element in series-elastic model | Vol II, Ch 5 |
| Differential learning | Noise injection; covariance maintenance | Vol IV, Ch 2 |
| Fascial transmission | Off-diagonal entries in \(G(\mathbf{x})\) | Vol II, Ch 3 |
The Bosch framework is not a competitor to the mathematical treatment of The Geometry of Motion — it is a practitioner’s intuition for the same underlying dynamics. The athlete who “exploits passive dynamics” is implementing \(\mathbf{u} \to 0\) with \(f(\mathbf{x})\) doing the work. The coach who says “relax and let the club do its job” is advising a reduction in feedback gain to allow the drift to dominate.