Control Loop Tuning Problems

A mechanism that overshoots, oscillates, or never quite reaches its setpoint is almost always a tuning problem with a known cause. WPILib documents the recurring ones.

Integral windup#

If kI is too large, the integral term accumulates error during a big setpoint change until it exceeds the maximum control effort -- the mechanism blows past the target and takes a long time to settle. Fixes, in order of preference (these methods live on WPILib's PIDController):

1. Reduce kI, often to zero -- most FRC mechanisms need no integral term at all. Tune the feedforward model instead.
2. Reset the integrator outside a tolerance band with setIZone() -- when the position error exceeds IZone, the accumulated error resets to zero, disabling integral until you are close again.
3. Cap the integrator with setIntegratorRange().

controller.setIZone(5.0);              // only integrate within 5 units of target
controller.setIntegratorRange(-2, 2);  // clamp accumulated effort

Voltage sag#

Drawing current drops bus voltage, so a controller tuned at 13V behaves differently at 11V -- the mechanism slows as the battery droops. Fix it by commanding voltage rather than duty-cycle percent so output is consistent regardless of bus voltage:

motor.setVoltage(feedforward.calculate(setpointVelocity));

Using setVoltage() (or the controller's voltage-compensation feature) makes your gains battery-independent.

Actuator saturation#

If gains are so aggressive that the mechanism demands more torque or acceleration than the motor can deliver, it saturates at maximum output and can become unstable -- you can't tune your way past physics. The fix is mechanical: change the gearing or move to a higher-power motor (e.g. NEO -> NEO Vortex, Falcon 500 -> Kraken X60) to gain control authority.

Missing or wrong feedforward#

The single most common 'bad PID' is actually missing feedforward. A flywheel or drivetrain is feedforward-dominated; an arm needs kG that varies with the cosine of its angle; an elevator needs a constant kG. If you try to do the whole job with kP, you get sluggish tracking and steady-state error. Get kS/kV/kA from a SysId characterization run, add kG for gravity-loaded mechanisms, and let P handle only the residual.

A disciplined process#

Start from zero gains. Add feedforward until the mechanism roughly follows commands open-loop. Then raise kP until tracking is crisp; add a little kD only if it overshoots; reach for kI last and rarely. Throughout, watch the measured-vs-setpoint plot in AdvantageScope -- oscillation means too much kP or noisy kD; lag means too little kP or missing FF; steady offset means missing kG or FF. Change one gain at a time and re-plot.