Use cases · Mobility · autonomy
Sign off an autonomy controller as safe against real disturbances
A disturbance model measured on a 128-qubit processor was the only one that certified a controller safe at its operating point.
Impact
A safety verdict you can defend
Baseline: the standard white-noise disturbance model: 39.5% safe episodes
Condition: 95.6% safe episodes with the measured disturbance model, same controller, same operating point
The problem
A controller tested against textbook white noise can pass and still fail against the disturbances it meets in service. A safety sign-off is only as good as the disturbance model behind it.
Our approach
We measured a disturbance model on a 128-qubit quantum processor and used it, alongside the standard white-noise model, as the input to a safety monitor that evaluated the controller at its operating point in simulation.
The evidence
- The measured disturbance model was the only one under which the controller was certified safe at its operating point.
- 95.6% safe episodes, against 39.5% under white noise.
- Not claimed: a learning-speed or alignment advantage, or any gain attributed to the device in training.
Status: Quantum hardware + Simulation. Disturbance model measured on quantum hardware; the safety evaluation runs in simulation.