Feed-forward Control

Robotics / control / pid

---

What is feed-forward control

Feedback control reacts to error after it appears. Feed-forward control adds a command from a model of the system before waiting for error.

For example, if a robot arm joint must hold a load against gravity, the controller already knows that some torque is needed just to hold the arm still. Instead of waiting for the joint to fall and then correcting the error, the controller can add the gravity compensation directly.

control = feedback + feed_forward

Feed-forward does not replace feedback. It helps the controller start closer to the right command. Feedback still corrects model errors, friction changes, disturbances, and sensor noise.

Simple idea

In this example the plant is a vertical mass:

mass * acceleration = force - mass * gravity - damping * velocity

The feedback-only controller uses position error to create force. That means it must first see an error before it can push hard enough to fight gravity.

The feed-forward controller adds the known gravity force:

feed_forward = mass * gravity
control = feedback + feed_forward

Now the feedback term only has to move the mass and correct small errors. This usually gives a faster and more stable response.

flowchart LR
    target["target"] --> feedback["feedback controller"]
    model["system model"] --> ff["feed-forward"]
    feedback --> sum(("Σ"))
    ff --> sum
    sum --> plant["plant"]
    plant --> output["output"]
    output --> feedback

---

Demo

Run the interactive demo:

python3 docs/Robotics/control/feed_forward/code/feed_forward_demo.py

Use the Feedback only button to run the controller without gravity compensation. Use the Feed-forward button to run the same controller with mass * gravity added to the command. Reset clears the graph.

The plot compares:

• position: how quickly the mass reaches the target height
• velocity: how much motion remains while settling
• force: the controller command, with mg marked as the force needed to hold the mass against gravity

The feed-forward case reaches the target because it starts with the force needed to cancel gravity. The feedback-only case is slower and has steady-state error because the controller must create gravity compensation indirectly through position error.