optimal control

Optimal control is the art of choosing, out of all the possible ways a robot could move from here to its goal, the one that is best by some clear measure you decide ahead of time. Imagine driving across town. There are countless routes, but you might want the one that uses the least fuel, or arrives soonest, or brakes the gentlest. Optimal control turns that vague wish into a precise number — a cost — and then hunts for the exact sequence of steering, gas, and braking that makes that number as small as possible.

The heart of it is the cost functional: a single running tally that adds up everything you care about over the whole journey, not just at one instant. Maybe each second of lateness adds a little cost, each hard jerk adds a little more, each drop of fuel a little more still. Because the tally stretches across the entire trajectory — the full path through time — the best choice now depends on what happens later. A move that looks wasteful this second might be the only way to avoid a far costlier scramble down the road, and optimal control accounts for that whole future.

This is the big idea that many famous controllers are special cases of. When the system is simple (linear) and the cost is written as squares, optimal control collapses neatly into the linear-quadratic regulator. When you must also respect hard limits — a motor that can only push so hard, a joint that cannot bend past a stop — and you re-solve the problem again and again as you go, you get model predictive control. Optimal control is the parent idea; those are its well-behaved children.