Skip to main content
Measures the torque at which a stationary joint first moves — its breakaway or static friction — in both directions, and compares it to the sliding Coulomb torque fc the friction model already carries. tune.friction cannot answer this: its model fc·tanh(0.1·k·v) + fv·v is monotonic in velocity and has no static/kinetic distinction, and its sweep bottoms out around 0.13 rad/s, above the creep speeds where a geared joint stick-slips. Stick-slip exists precisely because breakaway exceeds sliding friction — a joint tracking a slow target sticks until kp·err covers the gap, lets go, overshoots and re-sticks. On the X8-P20 shoulders that is a 2 Hz, ~0.5° stair pattern during slow extended-reach moves. This command measures the gap directly, with no velocity signal at all. Method, per pose. The arm is homed and the sweep-safety clearances applied; the test joint ramps to the pose under impedance control, then drops to kp = 0 with a small kd so only the gravity feed-forward holds it. That feed-forward is trimmed until the joint stands still (a joint sits still for any trim inside its stiction band, so this converges quickly — and without it a loaded pose is unmeasurable, since a 2 % gravity-model error at 15 Nm outweighs the whole breakaway torque). An extra torque is then ramped up and down in a triangle; the first motion past --move-deg is the release, the extra torque at that instant is recorded, and the joint is caught under kp and returned to the pose. Peaks escalate over attempts and stop at the first release, so the joint never sees more torque than it took to move it. What comes out. Releases in both directions split into a symmetric half — F_static, the friction — and an antisymmetric half, the standing torque the trimmed feed-forward still missed. The report gives F_static / fc, the predicted stick-slip stair (F_static − fc) / kp in degrees (compare it to the stairs a slow tune.motion replay shows), and the ceiling for the joint’s stiction_gain (F_static / fc − 1): compensation that exceeds the friction it compensates hunts around the target at rest, so start at 60–80 % of that ceiling.
The joint under test spends most of the run with no position spring — only a trimmed gravity feed-forward and light damping hold it. Keep the workspace around the arm clear and stay within reach of the e-stop, as for every tuning probe.