Digital Twins

A digital twin is a live simulation of a specific real machine, kept in sync with it through a stream of sensor data. Because the twin knows what should happen, a difference between twin and reality is a signal: a heavier payload, a worn gearbox, a collision. Twins are also used to test changes safely before sending them to the real robot.

Alarm threshold Sync delay from the real arm

Predict and compare

The twin runs the same commands through a model of the robot. The residual $r$, the gap between measured and predicted joint angles, should stay near the noise level. A residual above a threshold means the real robot is no longer behaving like the model:

$$r(t) = \big\lVert \mathbf{q}_{\text{real}}(t) - \mathbf{q}_{\text{twin}}(t) \big\rVert \qquad \text{alarm if } r > r_{\text{th}}$$

What to notice

Normally the twin tracks the real arm and the residual stays small. Add a payload: the real arm sags and lags, the residual jumps, and the alarm fires. Calibrating, updating the twin's mass to match, brings the residual back down, now with a model that matches reality again. A threshold too low gives false alarms from noise and sync delay. Twins built in Isaac Sim, Gazebo, or vendor tools are increasingly used to monitor robot fleets and to try new programs before running them for real.