Welcome to ControlMind
A home for essays and interactive simulations on control systems, robotics, and physical AI.
Why ControlMind
For most of the last decade, AI has lived in the cloud. It predicted the next word, ranked the next ad, and sorted the next image. The next decade is different. Intelligence is stepping out of the browser and into the world — onto sidewalks, factory floors, kitchens, and orbits. That world runs on physics, feedback, and delay. It does not care what your loss function converged to.
ControlMind is a place to think carefully about that transition.
What we'll cover
- Control systems — the oldest branch of engineering AI, built on stability, sensitivity, and feedback. Still the sharpest tool we have for reasoning about systems that move.
- Robotics — where algorithms meet actuators, and where clever code runs face-first into friction.
- Physical AI — learned controllers, world models, diffusion policies, VLAs. What works, what fails, and why.
Every essay ships with something you can poke at: a plot, a derivation you can re-run, or a simulation you can break. If you only remember the math, you didn't really understand it.
The first simulation
The inverted pendulum is the "hello world" of control. A cart on a track, a pole balancing on top, and a controller trying to keep it vertical. Start with the PID gains at zero — the pole falls. Turn them up, and a small miracle happens: the system becomes stable.
The controller in that demo is a textbook PID:
where is the angle error. Three numbers, and a linearization around the upright equilibrium, are enough to hold the pole up indefinitely — at least until you push the cart too hard.
We'll use that simulation across several upcoming posts to build intuition for pole placement, LQR, and eventually learned controllers trained in the same environment.
Follow along
New posts land here first. If you read on Medium, the mirror gets a canonical link back to this site — same words, same plots, but the interactive pieces only live here.
Welcome. Let's build.