In Layman’s Terms: Hybrid Tangent Spaces
A golfer can deliver almost the same club speed twice and still produce different shots. The contact point, face orientation, shaft state and timing may differ. A walking robot can reach almost the same leg posture twice and have different outcomes because one foot touches down earlier.
To understand these differences, we need to follow both the motion and the changes that contact introduces. A hybrid model describes smooth motion within a mode and rules for switching between modes. Its most useful lesson is that the result of a small change depends on what changes at contact and when contact happens.
This explanation accompanies the full derivations and executable example and the published critiques and responses. It develops the interpretation; the reference supplies the assumptions and calculations.
Motion, Contact, and Model Resolution
Imagine a ball falling toward the floor. During flight, its height and velocity follow a familiar gravity-driven path. During contact, the ball and floor deform, exert forces and exchange energy. The ball may then rebound.
We can describe that contact in two useful ways. A compliant model follows deformation and force over a short interval. An ideal rigid model replaces that interval with an instantaneous velocity change. The ball’s position remains continuous, while its velocity jumps in the idealization.
The rigid model has not proved that real contact takes zero time. It has chosen a resolution. It can tell us about momentum exchange without determining the peak force or the deformation history. If those details are the question, a more resolved contact model is needed.
The same distinction applies to a club striking a ball or a robot foot landing. A fast event is not a reason to discard physics; it is a reason to choose which physics must be resolved for the desired output.
Three Rules Define a Simple Hybrid Model
The Mode: What Rules Apply Now?
A mode is a particular set of dynamics. A hopping robot might have a flight mode and a supported mode. In flight it cannot push against the ground. During support it can exchange force with the ground, subject to contact and friction limits.
For a walking robot, one leg being in swing does not mean the entire robot is airborne. Its other leg may provide support. Useful whole-body modes distinguish left support, right support, double support and flight.
The Guard: What Triggers a Transition?
A guard is the condition that activates a switch. For a falling ball, it can be “height reaches the floor while moving downward.” The direction matters: the same height while moving upward should not trigger another bounce immediately.
Other transitions may depend on loss of support force or a controller threshold. If several conditions can trigger together, the model must explain how they are resolved. Simply writing several “if” statements can leave the outcome dependent on an accidental order in the code.
The Reset: What Changes at the Event?
The reset is the physical or modeled transition rule. In a simple bounce, a ball arriving at \(-5\) m/s might leave at \(+4\) m/s. The factor \(0.8\) is a chosen coefficient of restitution: the rebound normal speed divided by the incoming normal speed.
For a fixed floor, no air resistance and this constant coefficient, the next bounce height is \(0.8^2=0.64\) of the preceding height. That does not make \(0.8\) a universal value for rubber balls. Restitution depends on the objects, contact conditions and model.
At a robot foot strike, several body velocities can change together. Whether the foot stops, slips or rebounds depends on the impulse law and frictional feasibility. It cannot be decided from one joint’s speed alone.
Why the Timing Correction Matters
Consider two almost identical falling balls. One starts a little higher. It reaches the floor a little later.
At the moment the first ball has already bounced, the second may still be falling. If we compare their states at a common later time, the difference reflects several effects: the extra fall, a different incoming speed, the rebound rule, and less time since the later bounce.
Multiplying the incoming speed difference by the rebound coefficient captures only part of that story. It leaves out the changed timing.
A saltation matrix is the local sensitivity map that combines the reset’s effect with this timing correction. It acts on a small variation between trajectories. It does not replace the reset that acts on the actual ball.
This is the essential distinction:
- The reset tells us what happens to the nominal state at contact.
- The saltation matrix tells us how nearby trajectories differ after accounting for their different contact times.
A local sensitivity is a derivative. It predicts the leading effect of a sufficiently small change under stated assumptions. It does not make every finite change exactly linear.
Losing Energy Can Still Increase Sensitivity
A softer bounce loses more energy. It is tempting to conclude that it also suppresses every error. Those are different claims.
An energy balance follows one trajectory. Sensitivity compares nearby trajectories. A small height difference can produce a contact-time difference, which then becomes a velocity difference at a common observation time.
In the reference’s manufactured ball example, the bounce reduces the normal speed by a factor of \(0.8\). Nevertheless, its full timing-aware map can amplify a suitably scaled perturbation by a factor of about \(3.7\) in one event. Height and speed are scaled separately so that meters are not added indiscriminately to meters per second.
There is no contradiction. Each ball loses energy, but the difference between the two balls can become larger in some measured direction.
For a walking gait, we therefore examine the whole stride: motion between contacts, every contact transition, and the return to the next step. A stable return does not rule out a large temporary deviation within the stride. For a golf swing, the relevant question may be the spread in impact location or launch direction, rather than whether every internal state difference shrinks.
What This Adds to the Golf-Swing Investigation
Before Contact: How Was the Incoming State Produced?
Body motion, actuator forces, gravity, inertial coupling, ground support and shaft behavior shape the arriving club. A change at the shoulder or wrist can affect several later quantities at once: head speed, face orientation, impact offset and event time.
This is why “more torque” and “more club speed” are not interchangeable statements. A force acts through the current configuration and changes the subsequent motion. Its eventual effect depends on the trajectory and the rest of the system.
Similarly, a pelvis or arm slowing down does not by itself show that its energy went into the club. To establish transfer, the technical model tracks forces, moments, their associated powers and changes in segment energy.
During Contact: What Is Exchanged?
A simple two-mass collision is a useful starting point. With stipulated masses of \(0.200\) kg and \(0.046\) kg, incoming club speed \(50\) m/s, a stationary ball and restitution \(0.78\), it predicts ball speed about \(72.36\) m/s and outgoing club speed about \(33.36\) m/s.
Those numbers follow from the chosen assumptions. They are not a measurement of a particular swing.
An actual clubhead can rotate, strike away from its center and interact with a deformable, spinning ball. The effective resistance at the contact point depends on mass, rotational inertia and geometry. Shaft and grip coupling may also matter at the resolution of the model.
For the fixed two-mass assumptions, increasing incoming club speed increases outgoing ball speed. This does not settle the best full swing when accuracy, effort, torque limits or other outcomes matter. Nor does it establish that reducing torque just before impact is beneficial. That is a separate intervention to investigate.
After Contact: Which Outcome Matters?
Launch speed, launch angle and spin feed into ball flight. A simple vacuum projectile formula leaves out the aerodynamic effects that matter to golf. Face angle is also not automatically the same as launch angle.
“Improve the swing” therefore needs an explicit output. Carry distance, lateral dispersion, ball speed, repeatability and physical effort can favor different decisions. A model should show how its proposed change affects those outputs rather than hiding them inside one vague performance claim.
Shaft Flexibility, Contact Timing, and Control Are Intertwined
Suppose we change shaft stiffness. The shaft’s deflection and vibration phase can change. The golfer may change grip motion or applied forces in response. The head may arrive with a different position, velocity and orientation, leading to different contact conditions.
An experiment that prescribes the same grip motion for two shafts answers a different question from one that prescribes the same forces. Letting a golfer adapt introduces another response. A result from one setup cannot simply be transferred to the others.
The same care is needed for a “zero-torque” experiment in a model. Setting a specified generalized torque channel to zero does not necessarily remove passive tissues, gravity, existing muscle activation, shaft forces or ground reactions. The intervention must say what changes and what is retained.
Timing adds another distinction. If an altered swing reaches the ball later, evaluating it at the original impact time may compare a pre-impact state with a post-impact one. Comparing each swing at its own impact can be useful, but it answers a different question from comparing both at one clock time. A careful model labels that choice.
Where the Local Prediction Can Break
The ordinary saltation calculation assumes an isolated, non-grazing event and nearby trajectories following the same local transition sequence.
If a ball barely brushes a surface, a small change can turn contact into a miss. If two feet land almost simultaneously, a small change may reverse their order. If a swing narrowly misses the ball, one derivative calculated around successful impact cannot summarize every nearby outcome.
These cases do not mean all modeling has failed. They mean we may need separate contact branches, a more resolved contact law, or simulations that represent several possible outcomes. Smoothly curved guards can still be differentiated; the problem is not curvature by itself.
A measured uncertainty distribution can also span several branches. A single mean and covariance may then hide the distinction between “hits” and “misses,” even if it describes each branch locally.
The Infinite-Bounce Puzzle
In an ideal constant-restitution ball model, every bounce is lower and the intervals between bounces form a geometric sequence. The model can predict infinitely many events accumulating in finite time.
That is called Zeno behavior. It exposes a missing continuation rule: after the accumulation, should the model switch to resting contact, resolve deformation, or use another description?
Saying that each ideal bounce takes zero time does not answer that question. It also does not remove event costs. The total energy lost can be finite, while a bookkeeping cost of one unit per bounce would be infinite.
A simulator that declares rest below a threshold has added a rule. The rule may be useful and physically sensible, but its effect should be checked as the threshold changes. A numerical warning about rapid events is not itself proof of an infinite physical sequence.
Choosing a Contact Model
| Question | What the Model Must Resolve |
|---|---|
| How does a small height change alter a later rebound? | Event time, reset and outgoing motion |
| What impulse changes the ball’s speed? | Momentum balance and an appropriate contact law |
| What peak force acts during contact? | Force history, deformation and contact duration |
| Can a foot maintain support without slipping? | Normal-force and friction feasibility |
| Which contacts should a robot use? | Candidate contacts, admissible sequences and an optimization formulation |
| Does a proposed swing change improve repeatability? | A declared intervention, uncertainty, output and independent measurements |
An explicit impact model, a compliant model and a contact-implicit optimization can each be appropriate. None automatically guarantees accurate forces, fast computation or robust performance.
In contact-implicit optimization, gap and normal contact force are constrained so that a separated surface cannot carry a compressive contact force. The optimizer must still satisfy dynamics, friction and other constraints. Its successful termination does not prove that its contact assumptions match the physical system.
How Readers Can Check the Argument
The reference provides a complete one-bounce program. It finds the impact time, applies the rebound rule, and compares the predicted sensitivity with two nearby simulations. Each nearby simulation finds its own impact time.
This is a useful verification because an incorrect timing treatment can produce a plausible trajectory while giving the wrong derivative. The program also rejects horizons outside the one-bounce case it actually implements.
For a more ambitious swing model, ask what was measured, what was selected, and what was predicted. A calculation should make its assumptions visible, distinguish numerical verification from physical validation, and test predictions on data beyond those used to tune it.
The central connection is straightforward: forces shape the approach, contact transforms the incoming state, and timing changes the effect of a perturbation. Following all three lets us investigate how a swing works without treating one appealing mechanism as an explanation of every outcome.
Citation
@online{untitled,
author = {},
title = {In {Layman’s} {Terms:} {Hybrid} {Tangent} {Spaces}},
url = {https://affinedrift.com/articles/tangent-hyperplane-articles/Advanced/Hybrid_Tangent_LAYMAN.html},
langid = {en}
}