Characterizing Any Mechanism with SysId

Why characterize?#

Guessing PID gains is slow and fragile. Elite teams instead measure their mechanism's physics. The permanent-magnet DC motors in FRC closely obey the voltage-balance equation:

V = kS * sgn(v) + kV * v + kA * a (add + kG for gravity-loaded systems)

SysId, included with the WPILib installer, drives the mechanism through controlled test routines and regresses these constants from logged voltage/velocity/position data.

The four constants#

• kS (static): voltage to overcome friction/stiction. Nearly impossible to model, must be measured.
• kV (velocity): voltage to hold a given velocity, the dominant term.
• kA (acceleration): voltage to produce a given acceleration; matters during fast moves.
• kG (gravity): for arms/elevators, the voltage to counter gravity (cosine-scaled for arms).

The test routines#

SysId runs two kinds of tests in both directions:

1. Quasistatic: voltage ramps slowly so acceleration ≈ 0, isolating kS and kV.
2. Dynamic (step): a sudden voltage step, exciting acceleration to extract kA.

You wire your subsystem into a SysId routine, run quasistatic-forward, quasistatic-reverse, dynamic-forward, dynamic-reverse, then analyze the log.

Deploying the gains#

Drop the results into the matching feedforward (note the constructor order kS, kG, kV, kA for the gravity classes):

// Drive / flywheel
var ff = new SimpleMotorFeedforward(kS, kV, kA);
// Arm (kG cosine-compensated internally)
var armFf = new ArmFeedforward(kS, kG, kV, kA);
// Elevator (constant kG)
var elevFf = new ElevatorFeedforward(kS, kG, kV, kA);

Then add a modest kP for disturbance rejection. With good feedforward, kP does very little work, which is exactly what you want.

Per-side / per-module matters#

Characterize each drive side (tank) or each module (swerve) separately. Mismatched kV across sides is the textbook cause of a robot veering; distinct, measured gains fix it.

Pitfalls#

• Garbage in, garbage out: ensure encoder conversion factors are correct before testing, or every gain is scaled wrong.
• Run on the real surface with real mass; characterizing a bare gearbox gives gains that don't hold the loaded mechanism.
• Respect mechanism limits, set soft limits so a dynamic step doesn't crash an arm into a hard stop.

The payoff#

A characterized mechanism tracks motion profiles accurately, recovers from disturbances predictably, and behaves the same on Saturday as it did in your shop, the foundation every advanced technique below builds on.