Subsystem Decomposition: COM · Base · End-Effector

The wiki’s subsystem axis. The free-flying coordinated controller acts on three distinct
subsystems
; this page names them, defines the axis that tags them, and tabulates which failure
mechanism owns which subsystem
— the attribution discipline that
error_floor.md §4 demands.

Statement

The circumcentroidal coordinate transform (notation.md) sends
the full coupled dynamics into transformed coordinates
with a block-diagonal
inertia
(circumcentroidal_decoupling). That block structure is the
taxonomy — the three subsystems the thesis tracks (errors ;
inputs ):

  • COM (com) — the decoupled loop , closed by
    the SPD impedance .
    Inertially free (); sees no Coriolis coupling.
  • Base attitude (base) — , the upper block of the coupled
    circumcentroidal subsystem .
  • End-effector (ee) — , the lower block of that same
    .

base and ee are not independent: they share the reduced inertia and the
CoM-coupling Coriolis , forming the coupled attitude+EE state
. Only com truly decouples.
Frames are in notation.md; the frame schematic is
Giordano Fig. 1 (library/coordinated_control/coordinated_control.md).

The subsystem axis

This decomposition is the wiki’s subsystem axis — a list-valued typed field (closed vocab
com | base | ee; omit ⇒ subsystem-agnostic). A page carries every subsystem its claim touches:
a CoM-loop result is [com]; a circumcentroidal-block mechanism is [base, ee]; a whole-controller
overview is [com, base, ee]. Ask “what touches the COM?” with Dataview
WHERE contains(subsystem, "com"). The axis is the join key across the three layers: the concept
wiki (here), the code wiki (sims_implementation_map), and
the logged evidence (evidence_map).

Failure modes are subsystem-specific (the attribution discipline)

Decompose before diagnosing: each subsystem has its own singular set and its own failure
mechanism, and conflating them mis-assigns cause. Distilled from
error_floor.md §4:

SubsystemBlockTrigger / singular setCharacteristic failureMechanism
com decouplednone intrinsic ()near-singular spikes in open — likely forward-Euler if develops a fast mode; not Tikhonov
base, ee ( rank loss)velocity-reconstruction blow-up; period-2 rippleforward-Euler on (discretization_stability_omega_b); Tikhonov / derate act on this block (singularity_threshold_cascade)
control mapquaternion singular set (distinct from )error-floor premise loses invertibilitysteady_state_error_floor

The discipline. Name the subsystem and its block before attributing a fix. Tikhonov
regularization acts on the base+ee -solve — it does not act on the
decoupled CoM loop.
Whether near-singular CoM spikes are the same forward-Euler artifact
or a genuinely separate CoM-loop effect is an open question, decided by the -independent
lag-1 autocorrelation of the logged ( at
flipping to at ⇒ forward-Euler). Evidence:
evidence_map. Do not write “Tikhonov causes the COM spikes” until
that test is run — this is exactly the mis-attribution
error_floor.md §4 flags.

Assumptions

  • Nonsingular invertible: the decoupling
    holds on the singularity-free region (singularity_threshold_cascade).
  • (nonredundant) for the clean / split. The redundant case adds a
    self-motion / null-space channel (frozen below ) — a redundant DOF outside this
    three-subsystem taxonomy.
  • Hat-vs-breve is load-bearing: full , reduced
    (terminology.md, circumcentroidal_decoupling).

Proof sketch

The block-diagonal that
licenses the split is derived in
dynamics_modifications_7dof.md and
circumcentroidal_decoupling; the subsystem-specific failure-mode
separation (CoM forward-Euler vs base+EE blow-up) is established in
error_floor.md §1–4 and
omega_b_forward_euler_instability.md.

This page is the taxonomic synthesis, not a new theorem: it names the three subsystems the proven
decomposition produces and records which mechanism owns which block.

Source / provenance

  • Literature: giordano2019coordinated — the
    split (eq 21) producing the decoupled CoM + coupled attitude+EE blocks; Fig. 1 frames.
  • Ours: the subsystem axis and the failure-mode-by-subsystem attribution table — the diagnostic
    discipline that the CoM-spike attribution question motivated.

Caveats

  • base and ee are a coupled pair, not independent subsystems; [base, ee] reflects the shared
    block. Tagging a circumcentroidal-block claim with only one of them understates the
    coupling.
  • The CoM-spike attribution is unresolved; this page records the open question and the deciding
    test, it does not assert the answer.
  • Distinct singular sets: (circumcentroidal, base+ee) the quaternion singular set
    of (the error-floor domain). Do not merge them.

Implementation (sims wiki)

External — into the code wiki via the sims_wiki/ symlink (resolves in Obsidian, not GitHub).