A control loop that closes on the machine
A control loop measures, corrects, and repeats on a fixed beat. A late correction is a wrong one, which is why the loop stays on the machine and why the first E31 design is a real-time controller.
· 4 min read
IN SHORT
Real-time control stays on the machine because a control loop must close on time, every time, and a network link cannot promise that. The first BramHA design is a real-time controller with a root of trust, being designed so that the loop closes on the chip. It may run a vendor's control loop; the safety case stays the vendor's. No chip exists.
Every moving machine runs a control loop. It is the cycle that keeps a joint where it should be and moves it where it should go. The first design in the BramHA program is a chip for that cycle: a real-time controller with a root of trust.
The program is in development and no chip exists. This article explains what a control loop is, why it stays on the machine, and what a real-time controller is being designed to do.
What a control loop is
A controller reads where a joint is. It compares that with where the joint should be, computes a correction, and commands the drive. Then it does the same thing again, and again, on a fixed beat.
The beat is the point. Each correction is based on a measurement, and a measurement goes stale. A correction that arrives late acts on a position the joint has already left. Late enough, and the motion becomes rough or unstable.
So in control, a late answer is a wrong answer. A loop is said to close when the measurement, the calculation, and the command complete within the beat. A real-time system is one that can promise that they will.
Why the loop does not cross a network
A facility's network is built to reach every machine and carry a great deal of traffic. It is not built to guarantee a beat. A link can be slow for a moment. It can drop. A pool that serves many machines answers them in turn.
None of that is a fault in the network. It is why the loop does not depend on it.
This is the line Element 31 draws for the whole family. A Czar Grid serves perception, decisions, and adaptation over the site's network. It is not a controller and does not sit inside any real-time or safety loop. The pool can tell a machine what it is looking at and what to do next. The machine does the doing.
If the link goes quiet, the loop keeps closing. The joint is still held, and the machine can still stop.
What a real-time controller is being designed to do
The controller is being designed so that the loop closes on the chip. In each cycle it is intended to read its inputs, run the control calculation, and write its outputs, and to finish inside the beat every time.
That asks for a particular kind of design. What matters is the worst case, not the typical case. A processor that is very fast on average but occasionally slow is a poor controller. A modest processor whose slowest cycle is known and bounded is a good one.
This is why the job does not need the leading edge of chip manufacturing. Predictable timing comes from how a chip is designed, not from how small its transistors are. Determinism over density sets out what a control job asks of a chip.
The control loop itself is the robot maker's. The chip may run a vendor's control loop; it does not supply one. The control law, the tuning, and the safety case built around them remain the vendor's.
What it is not
The first design is not a vision processor, and it is not an AI accelerator. It is not intended to run a large model or to decide what the machine should do next. That work is served by the pool in the building.
It is not a safety system. No safety or functional-safety certification is claimed for it, and none should be read into the word controller. Safety, stops, and motion stay on the machine as its vendor built them.
And it is small. A small first design does little. The program accepts that in exchange for a design that can be specified, verified, and understood in full.
Where the design stands
No chip has been taped out. The specification of the first design is not published, so there are no figures for timing, power, or anything else. The process and the foundry have not been announced.
What can be said is the intent: a controller whose loop closes on the machine, with a root of trust beside it so that the machine can prove what it is. The program's page at /program describes how the design is being built and what is not done.
Questions this raises
- Why can a fast network not close the loop?
- Speed is not the issue. A control loop needs a result on a fixed beat every cycle, and a network link can be slow for a moment or drop. A loop that depends on the link is only as dependable as the link.
- Does the chip supply the control software?
- No. The controller is being designed to run a robot maker's control loop. The control law, its tuning, and the safety case built around it are the vendor's.
- How fast is the controller?
- There are no performance figures. No chip has been taped out, and the specification of the first design is not published.
KEEP READING