Note: The following systems describe classic control schemes. For learning based approaches see Robotic control with Deep Reinforcement Learning
RealPars: What is a PID Controller? (video)
Definition: Proportional, Integral, Derivative controller
Terms:
- Process Variable (PV)
- measured/sensor value
- Set Point (SV/SP)
- target value
PID controllers can work as on-off controllers to regulate between two acceptable values around the set-point. Tightening the min/max band and feedback time can get the value closer to optimal/set-point.
RealPars: How to Tune a PID Controller (video)
PID Parameters:
- Error (E)
- Difference between setpoint and process variable (SP-PV)
- Proportional (P)
Each parameter of P, I, and D is multiplied by the error.
Most widely used tuning method: trial & error (lol)
Also known as:
- proportional gain
- gain
- Proportional band (when expressed as a percent)
- K_p
Determines how fast the system responds (rise time?).
Higher gain (lower proportional band) means more sensitive/less stable loops.
AKA:
- reset
- K_i
- T_i = (1/K_i)
can be measured in:
- repeats/sec
- repeats/min
- sec/repeat
- min/repeat
sum of all PV values recorded from the signal within a set time
Usually measured in seconds or minutes.
Predicting change.
Tuning this parameter means setting how far ahead to try predict change.
PV must be a very clean signal for derivative to work properly. .: PI controllers are much more common
- Increase Gain (P) by double until oscillation occurs (disable I,D)
- Set Gain at 50% of value at which oscillation started
- Increase I by double until oscillation occurs in PV
- Set I at 50% of value at which oscillation started
- Fine-tune parameters further by small increments
TODO Deeper notes from link below Source: Wikipedia PID article