I am reading about feedforward in control system. And in this lecture, slide 49, the author says that feedforward is common in human systems. Example, walking, playing basketball, driving a car. Could you explain how feedforward is used in walking?
- Anybody can ask a question
- Anybody can answer
- The best answers are voted up and rise to the top
A good example is driving a car around a curve.
In a pure feedback system, you go straight until the error signal tells you that you're not in the center of the lane anymore, then adjust the steering to compensate.
Humans look ahead and see that a curve is coming up, and actually turn the wheel a little before entering the curve. This pre-compensates for the lag in the car responding.
The other examples you cite are too complex to be useful in illustrating just the feed-forward part of the behavior. There is a lot more to riding a skateboard, for example, than a simple control loop with feed-forward.
I believe the author is making an analogy to human behavior, not human motion.
For example: I'm playing basketball, and I want to score. I know I probably need to dribble towards the basket, avoiding defenders, then throw the ball at the hoop. I know all of this without any feedback: it is a feed-forward system.
A few slides later (50), he writes:
That is, if I knew exactly what all the defenders would do (the model), I could score points in basketball using only a feed-forward system.
Of course that's not exactly true. But I do know a good deal about the model, especially if I'm a good player. I can probably anticipate what the defenders will do more or less, before I make the play. I'll pick my play accordingly.
On the next slide, the author then discusses adding feedback to deal with these unknowns, since most systems don't perfectly know the model.
This is in contrast to purely feedback systems (PID controllers). By incorporating a priori knowledge about the system, and incorporating feedforward into the system, accuracy can be improved and the disadvantages of feeback-only systems can be overcome.
In other words, a basketball player who could not formulate a strategy in advance would not be very good.
That said, from my (rudimentary) understanding of biology, there is a feed-forward component in motion also, with respect to the brain. Check out this TED talk: A robot that runs and swims like a salamander
There are plenty of examples.
If you walk forward, and you have no sense of touch and you're blind, as well as muted (and blocked of whatever other senses). Unless you have perfect knowledge of the world model, and of every single consequence a motor cortex action , you're bound to go off track the desired goal (moving forward).
Vision feedback is one thing that can assist you in detecting error (misplacing foot), and correcting the input. Or if you're daredevil, hearing can help you.
High brain areas in neocortex formulate plans, and feed these plans forward to lower brain areas and spinal cord, which implement them, presumably using inverse models of the motor systems that would be implementing them to generate low level commands. There is, of course, local feedback to keep these plans going right in the presence of surprises and other perturbations.
When I lecture about reflexes (in the context of a physiology course taught to engineers), I take a huge Neuroscience textbook which I've hollowed out to weigh much less than one would think looking at the book, and I toss it to a student. The student, of course, handles it just fine, using spinal reflexes to keep from mishandling the book, none of which involves cortex.
Slightly aside, there is an incredible amount of local feedback at the level of spinal cord, as can be seen in this video of a decerebrate cat walking on a treadmill changing gait as the treadmill speeds up: https://youtu.be/wPiLLplofYw