A grip that stays in place.

A frozen comparison of four hold controllers on eight new simulated scenes. Holding the object and keeping the tool in place are measured together.

32 scheduled attempts. Every failure stays in the denominator. Authored controls using simulator state; no model training or hardware qualification.

Five-second outcomes

Controller	Admitted	Native grip	Native stationary grip	Active stationary grip
measured joint	4/8	0/8	0/8	0/8
issued joint	4/8	4/8	0/8	0/8
cartesian	4/8	4/8	0/8	0/8
incremental	4/8	1/8	1/8	0/8

The primary endpoint requires continuous grip plus TCP motion within 4 mm and 0.05 rad. “Active stationary grip” additionally requires the controller to remain qualified. A post-refusal fallback can retain an object; that does not establish successful active regulation.

Every attempt

measured joint · all eight attempts

Scene	Valid	Grip	Stationary grip	Active stationary grip	TCP excursion	Reason
m0475-f78-p0	yes	no	no	no	1.037 mm	native_predicate_lost
m0475-f78-p1	yes	no	no	no	1.724 mm	native_predicate_lost
m0475-f52-p0	yes	no	no	no	unavailable	no_admission
m0475-f52-p1	yes	no	no	no	unavailable	no_admission
m0525-f78-p0	yes	no	no	no	2.100 mm	native_predicate_lost
m0525-f78-p1	yes	no	no	no	2.216 mm	native_predicate_lost
m0525-f52-p0	yes	no	no	no	unavailable	no_admission
m0525-f52-p1	yes	no	no	no	unavailable	no_admission

issued joint · all eight attempts

Scene	Valid	Grip	Stationary grip	Active stationary grip	TCP excursion	Reason
m0475-f78-p0	yes	yes	no	no	4.797 mm	tcp_stationarity_lost
m0475-f78-p1	yes	yes	no	no	4.765 mm	tcp_stationarity_lost
m0475-f52-p0	yes	no	no	no	unavailable	no_admission
m0475-f52-p1	yes	no	no	no	unavailable	no_admission
m0525-f78-p0	yes	yes	no	no	4.704 mm	tcp_stationarity_lost
m0525-f78-p1	yes	yes	no	no	4.673 mm	tcp_stationarity_lost
m0525-f52-p0	yes	no	no	no	unavailable	no_admission
m0525-f52-p1	yes	no	no	no	unavailable	no_admission

cartesian · all eight attempts

Scene	Valid	Grip	Stationary grip	Active stationary grip	TCP excursion	Reason
m0475-f78-p0	yes	yes	no	no	4.207 mm	tcp_stationarity_lost
m0475-f78-p1	yes	yes	no	no	4.208 mm	tcp_stationarity_lost
m0475-f52-p0	yes	no	no	no	unavailable	no_admission
m0475-f52-p1	yes	no	no	no	unavailable	no_admission
m0525-f78-p0	yes	yes	no	no	4.260 mm	tcp_stationarity_lost
m0525-f78-p1	yes	yes	no	no	4.242 mm	tcp_stationarity_lost
m0525-f52-p0	yes	no	no	no	unavailable	no_admission
m0525-f52-p1	yes	no	no	no	unavailable	no_admission

incremental · all eight attempts

Scene	Valid	Grip	Stationary grip	Active stationary grip	TCP excursion	Reason
m0475-f78-p0	yes	yes	yes	no	1.580 mm	success
m0475-f78-p1	yes	no	no	no	1.582 mm	native_predicate_lost
m0475-f52-p0	yes	no	no	no	unavailable	no_admission
m0475-f52-p1	yes	no	no	no	unavailable	no_admission
m0525-f78-p0	yes	no	no	no	1.634 mm	native_predicate_lost
m0525-f78-p1	yes	no	no	no	1.638 mm	native_predicate_lost
m0525-f52-p0	yes	no	no	no	unavailable	no_admission
m0525-f52-p1	yes	no	no	no	unavailable	no_admission

What changed

The new incremental candidate accumulates the current pose-error step around the target issued before admission. It retains the jaw request, limits each correction update to 0.002 rad and refuses a proposed total above 0.03 rad. Candidate measurements outside the native joint margin are refused before solving. The three previous controllers remain unchanged.

These are nearby deterministic development cells, not independent random trials or a real-world reliability estimate. The study does not test transport, placement, videos, cameras or cross-arm transfer. Timing in the JSON is shared-host offline computation, not sensor latency.