Feedforward control computes the command a system should need from a model, before any error appears. Feedback then only corrects what the model gets wrong. Most real motor controllers use both. It complements a Feedback Control System.
Feedback Only Versus Feedforward Plus Feedback
Feedback only:
- the controller must first see an error before it acts
- to respond fast it needs high gains, which can cause oscillation
Feedforward plus feedback:
- the model gives most of the command at once
- the feedback controller only trims small errors, so its gains can be small and stable
Simple Speed Example
For a motor whose speed is roughly proportional to duty cycle:
- is the measured speed at full duty
- limiting the PID output (for example to of full duty) keeps the feedback as a small correction
Common Feedforward Terms
- velocity feedforward: command proportional to the target speed
- acceleration feedforward: extra torque for the target acceleration, from the inertia
- gravity compensation in robot arms: the torque needed to hold each joint against gravity
- friction compensation: a minimum command to overcome static friction
Limits
- feedforward needs a reasonable model, and a wrong model gives a wrong command
- it cannot reject disturbances it does not know about, which is why feedback is still needed