Lab 1 / 15 · Interactive · real-time
PID Tuning Lab
Drag three gains and watch one control curve go from sluggish to oscillating and back to stable.
Controller parameters
Readouts
The plant is first-order inertia plus 40 ms of pure delay, modeling a real joint (motor fast, gearbox and communication lag behind). Pure delay is the root of overshoot.
What problem this lab solves
PID is the one thing in robotics that every role touches — the algorithm engineer uses it for outer loops, the embedded engineer for current loops, the mechanical engineer to explain why a joint shakes. Textbooks teach three letters; real machines teach three gains fighting each other.
How to read the figure
The horizontal axis is time, the vertical axis is position. The gray line is the target, the red dot is the measured value. Watch four numbers: overshoot (how far past the target), rise time (how long to reach 90%), settling time (how long to stay within ±2%), and steady-state error (how far off it ends). Higher Kp → faster rise, more overshoot; higher Kd → damps overshoot but amplifies noise; Ki accumulates → kills steady-state error, but slows things down and causes low-frequency wobble.
Hands-on
Push Kp from 1 to 12 and note the critical value where overshoot first exceeds 20%. Then fix Kp and raise only Kd: can it push overshoot back down, and what does it cost (watch the spikes on the curve)? This is the real answer to «how do you tune PID» in interviews.
Done playing — where next