Classical Control Is Not Dead: Why Robots Still Need Physics

Industrial robot working in foundry

Robotics is now very connected to AI. Every week there is new videos about humanoid robots, robot dogs, or some learning algorithm which can do a task better than before. It is exciting, and I also like this direction.

But sometime I feel the discussion become too simple. Like, if AI is strong enough then old control methods are not needed anymore.

I don’t think this is true.

A robot is not only a software. It has mass, motors, friction, delay, cables, gears, and sometimes contact with human or ground.

So even if the policy is learned by AI, the robot still live in physics. The body does not disappear.

Why classical control still matter

Classical control methods like PID, computed torque control, impedance control, or model predictive control are not just old formulas. They give us a way to use what we already know about the system.

For example, if we know the dynamics of a robotic arm, we should not ask a neural network to learn all of it from zero. It can learn, yes. But it may need many data, many failures, and maybe some unsafe actions too.

In real robots this is not always acceptable.

A simple robot dynamics equation is usually written like this:

M(q)\ddot q + C(q,\dot q)\dot q + G(q) = \tau

This equation is not the full reality, of course. But it is still useful. It says that the torque is related to inertia, velocity effects, gravity, and the motion we want.

For me this is the main point: physics give a structure before learning start.

Where AI can help

Learning based control is very useful when the model is not perfect. And the model is almost never perfect.

There is friction that we did not model well. There is uncertainty in mass. There is sensor noise. There is environment changes. Sometimes the robot touch something and the contact is hard to predict.

In these places, learning can add good correction.

This is why I like residual learning. The basic controller do the main job, and the learning policy only add the missing part.

Flowchart of classical controller with learning residual correction — One possible hybrid control idea: classical control gives the main torque, while learning adds a correction torque.

This is not “AI against control”. It is more like AI beside control.

Maybe this is less dramatic, but for engineering it make more sense.

A small connection to my own work

In my master thesis, I worked on a cable-driven lower-limb rehabilitation robot. I used computed torque control as the baseline, then added residual DDPG to reduce tracking error under disturbances and model uncertainty.

The important idea was not to remove the classical controller. I used it because it already understand part of the robot dynamics. The reinforcement learning part was added as compensation.

In simulation, this gave better tracking than only computed torque control.

So for me this subject is not only a trend. I saw in my own simulation that hybrid control can be more practical than choosing one side.

Why this matter more in robots than in normal AI

In text or image AI, a wrong output can be corrected. In robot control, a wrong command can move a motor, hit a limit, or push too much force.

This is why control design is not only about performance. It is also about safety and predictability.

FANUC welding robots

For industrial robots, rehabilitation robots, and mobile robots, we need systems that are not only smart in demo videos. They should also be stable, explainable, and tested under bad cases.

That is why simulation and digital twins are becoming very important too. But even simulation need physics. If the simulator is not physical enough, the learned behavior may fail when it go to the real robot.

My view

I don’t think classical control is dead. I think it is becoming part of a bigger stack.

Maybe the future robot controller will have a physics model, a classical feedback loop, a safety layer, and a learning policy together.

Some part are hand-designed. Some part are learned from data. The best result probably come from using both.

So the question is not:

should we use control or AI?

The better question is:

what should be modeled, what should be learned, and where we need safety boundaries?

That question is maybe less popular. But it is more real.

Image credits

classical-control-robot-foundry.jpg: Wikimedia Commons, Automation of foundry with robot.jpg, public domain.
classical-control-fanuc-welding-robots.jpg: Wikimedia Commons, FANUC 6-axis welding robots.jpg, CC BY 3.0, author Phasmatisnox.
residual-learning-control-flow.png: Simple diagram generated for this blog post.