Intelligent motion.
Defensible safety.

Start with the failure

The proposed method starts with the risk assessment and an explicit set of relevant faults. For each fault, estimate a three-dimensional reachable volume, including motion during detection and response, subsequent loss of stability, and payload movement where applicable. Combine these volumes by geometric union.

What needs to be established

Credible bounds depend on the robot's geometry, initial state, drive and brake behaviour, floor conditions, payload retention, and measured response times. A changing boundary would also require a validated means of enforcing it. This proposal is a research method, not a certified safety function or a substitute for applicable safeguarding requirements.

Engineering question: Which assumptions bound the motion, and what evidence supports each one?

Permission must remain justified

Continuous Intent Validation (CIV) is Tom Vardon's developing concept for continuously checking that measured behaviour remains consistent with predicted behaviour. Its residual is expressed as r = x_m − x_hat, where x_m is the measured state and x_hat is the predicted state.

A residual alone is insufficient. The design must define comparable state variables, measurement uncertainty, allowed deviations, timing limits, and a specified response when validation is lost. Completion validation provides permission to continue, subject to the checks remaining valid.

What needs to be established

Research must examine model error, sensor faults, common cause failures, independence, timing, and the reliability of the response path. No performance level or safety certification is claimed for CIV.

Engineering question: Can an incorrect action be detected and controlled before it becomes hazardous?

A test should answer a defined question

A useful fault study records the initial conditions, injected fault, observed behaviour, detection time, response time, and final state. Results must be assessed against criteria established before the test.

Simulation can explore combinations that are difficult to investigate physically. Its conclusions remain conditional on the model and its validation. Physical measurements, documented assumptions, and traceable results are needed to support application-specific decisions.

Engineering question: Does the evidence support the claimed behaviour across the relevant operating conditions?