Real-Time Control Loops: Why Update Rate Matters

The speed at which a controller recalculates its commands is its own critical spec, separate from raw processing power.

A control loop's update rate is how many times per second it recalculates and issues new commands. Balance and low-level joint control on a humanoid typically run at several hundred to a couple thousand updates per second; higher-level planning and vision-based decisions run much slower, often tens of times per second.

Why faster isn't always better

A faster loop can react to disturbances sooner, but every loop is bounded by how fast sensor data actually arrives and how much computation fits in that time budget. Running a loop faster than its sensor data can meaningfully update just adds computational cost without adding real information.

'Real-time' means predictable, not just fast

In control systems, 'real-time' specifically means a computation reliably finishes within a guaranteed time window, every single time — not merely that it's usually fast. A control loop that's fast on average but occasionally misses its deadline is often more dangerous than one that's consistently a bit slower, which is why safety-critical robot control software is built and tested around worst-case timing, not average timing.