Critique: Controllability-Drift Ratio (Formalism Overreach)

Critique and response context for Controllability-Drift Ratio (Formalism Overreach) in AffineDrift’s control-affine golf-swing framework.

Critique: Controllability-Drift Ratio (Formalism Overreach)

Summary of Concern

The article articles/controllability-drift-ratio.qmd invokes Lie Bracket analysis—a tool specifically designed for checking local accessibility in nonholonomic or underactuated systems—to describe the loss of control in a system that is modeled as fully actuated (3-link arm-forearm-club with 3 torques) but input-saturated.

The text claims that “As drift increases… \(\dim(\mathcal{V}(x)) \longrightarrow 1\)”. This is mathematically incorrect for a fully actuated system, where the rank of the control distribution (and thus the Lie algebra) remains constant. The loss of control is due to finite-time reachability bounds (saturation), not a collapse of the geometric structure of the manifold.

Location

  • Page: articles/controllability-drift-ratio.qmd
  • Section: 4. Nonlinear Controllability via Lie Brackets
  • Claims:
    • “control cannot access or excite these directions” (implies rank loss)
    • “\(\dim(\mathcal{V}(x)) \longrightarrow 1\)” (implies dimension is a continuous variable or collapses)
    • “Lie bracket analysis reveals the geometry of that loss”

Nature of the Issue

  • Formalism Overreach / Math Washing: Using advanced differential geometric control theory (Lie Brackets, Frobenius Theorem context) to describe a phenomenon (actuator saturation relative to drift) that is adequately explained by simple vector magnitude ratios.
  • Category Error: Conflating Local Accessibility (geometric existence of a path, \(Rank(Lie) = n\)) with Small-Time Local Controllability (STLC) under saturation (practical ability to move against drift).
  • Mathematical Incorrectness: For a fully actuated system (\(rank(g) = n\)), the Lie brackets are zero or linear combinations of \(g\), and the rank is always \(n\). It does not “approach 1”.

Why This Is a Problem

  • Credibility Risk: A control theorist reviewer will immediately spot that Lie Brackets are irrelevant for a fully actuated robot arm. It looks like the author is “throwing math” at the problem to make it look deeper than it is.
  • Misleading Intuition: It suggests that the directions of motion are lost. In reality, the directions are available, but the magnitude required to move in them exceeds the actuator limits. The “cone” narrows not because the manifold geometry changes, but because the “allowable control set” is small relative to the drift vector.

Evidence / References

  • Murray, Li, Sastry (1994): A Mathematical Introduction to Robotic Manipulation. Lie brackets are used to analyze nonholonomic constraints (rolling without slipping). A robot arm is holonomic.
  • Isidori (1995): Nonlinear Control Systems. Controllability rank conditions are discrete (full rank or not).
  • Article Text: Section 2 defines \(q \in \mathbb{R}^3\) and \(\tau \in \mathbb{R}^3\) (implied). Section 2.2 defines \(g(x)u = [0; M^{-1}\tau]\). Since \(M^{-1}\) is full rank, \(g(x)\) spans the acceleration space.

Severity

  • High: This is a fundamental misapplication of a core control-theoretic concept. It invalidates the “rigor” claimed in that section.

Suggested Remedies

  1. Remove Section 4 (“Nonlinear Controllability via Lie Brackets”) entirely if the system is fully actuated.
  2. Replace with “Reachability Analysis”: Frame the “Control Cone” in terms of the Hamiltonian \(H(x,p,u) = p^T (f(x) + g(x)u)\) and the reachable set boundary, rather than Lie Algebra rank.
  3. Correct the Dimensionality Claim: Instead of “dim -> 1”, state “The volume of the reachable set (relative to drift displacement) shrinks.”
  4. Clarify Actuation: If the wrist is considered passive (flexible hinge) in this specific derivation, state it clearly. But Section 2 says “q3: wrist/club hinge”, implying actuation.