Textbooks teach the structure of a control loop. They do not, in general, teach you the dozen-or-so techniques you’ll actually deploy when you discover that the structure isn’t enough. There’s a transport delay that breaks your beautiful PI design. There’s measurement noise that makes derivatives explode. There’s an integrator that winds up to 10× the actuator saturation limit and refuses to come back. There’s a resonance you don’t have time to design around.
The standard response, in practice, is one of twelve techniques — sometimes two or three of them stacked. Each technique has a clean explanation and well-understood implementation pseudocode. Knowing all of them is the difference between a control engineer who understands loops and one who ships them.
This manual is those twelve techniques, each with a live simulation you can drive directly.
What it covers
Twelve sections, every one with an interactive plant simulator that integrates in real time.
01 — The Smith Predictor. When transport delay dominates, classical feedback is blindfolded. The Smith trick gives the controller a delay-free model to act on. With caveat sheet about model mismatch.
02 — Two-Degree-of-Freedom PID. Why setpoint changes and disturbance rejection want different controller gains, and how to give them different gains without breaking the loop.
03 — Anti-Windup. Conditional integration, back-calculation, and clamping. Three patterns that turn a saturated integrator from a liability into a non-issue.
04 — Feedforward Control. Using measured (or modeled) disturbances directly, instead of waiting for the loop to feel them.
05 — Derivative Kick & Filtering. Why a clean derivative is impossible and what the standard tricks are — derivative on PV-only, two-pole filter, smooth setpoint transitions.
06 — Cascade Control. When you have a fast inner variable and a slow outer one, the right architecture is two loops, not a single complicated one. With tuning rules for the speed separation.
07 — Gain Scheduling. Looking up controller gains by operating point. The simple version, the smooth-interpolation version, and the well-posedness caveats.
08 — IMC / λ-Tuning. Internal Model Control as both a theory and a recipe for first-cut PID tuning when you have a process model.
09 — Disturbance Observer (DOB). Estimating the disturbance directly and canceling it at the input. The hidden gem that powers a lot of robotic high-performance control.
10 — Notch Filters for Resonance Suppression. A short detour into the second-order band-stop, depth-vs-width tradeoffs, and why notches usually beat detuning when the resonance is sharp.
11 — Bumpless Transfer. Switching between controllers (auto-to-manual, fault recovery) without injecting a step into the actuator.
12 — Reference Shaping & Input Shaping. Convolving the reference with a few well-placed impulses to suppress flexible-mode oscillation. The technique that quietly ships on every coordinate-measuring machine.
Plus glossary and rules-of-thumb pages at the end.
A note on visual style
This manual is rendered with a dark theme — near-black background, mint-green accent — rather than the warm-paper aesthetic of the rest of Autonomy. The dark theme is a deliberate choice for this kind of operational/engineering reference: think of it as a terminal window for a working engineer, rather than a magazine article.
Read it
← Back to Autonomy