Engineering guide
Identify how AI can affect a robot hazard
Connect AI outputs to physical actions and the evidence needed to trust them.
Reviewed 2026-11-04 · Engineering interpretation · authoring reviewed
Map the influence
For an original assessment worksheet, name the AI input, output, downstream command and physical hazard. Consider missed detection, incorrect motion, stale data and behavior outside tested conditions. Identify which protective function acts if the AI is wrong and whether that function depends on the same sensing, software or power supply.
State evidence limits
This proposed review asks for operational limits, representative failure tests, version control and a response to uncertainty. Model accuracy alone leaves the physical stop, timing and restart questions open. The three worked cases are a starting structure for those questions. NIOSH identifies unexpected contact and close human–robot interaction as continuing safety concerns.
Sources & editions
Robot Safety Forum original engineering notes
Initial publication edition 2026-11-04
Original AI influence and shared-dependency worksheet