Why the safety concept has to come first, not last

A working cell that cannot be switched on is not a contradiction; it is the single most common outcome of treating the safety concept as the last step rather than the first one, and it happens after the engineering is already finished.

A humanoid's own dimensions are the starting point of a risk assessment, not a footnote to it.

The A2 Ultra stands 169 cm and weighs 69 kg, close enough to an adult worker's own height and mass that the swept volume around it during a reach or a turn is not a small correction to a factory layout, it is a boundary that has to be drawn before the layout is finalised. A risk assessment done against a finished cell is checking whether that boundary happened to fit; done first, it decides where the cell's edges actually are.

The failure this produces is specific, not vague: a cell that works perfectly and cannot be switched on for weeks after it is built.

The engineering is finished, the pick-and-place routine runs cleanly in testing, and the safety sign-off then asks for a defined stop boundary, a named person who owns the decision to halt the machine, and a way to stop it that does not depend on the robot noticing anything on its own. None of those three is a hardware change; all three are decisions, and deciding them after the cell is physically built means retrofitting a boundary around equipment that was never laid out with one in mind.

A robot safety assessment and a humanoid risk assessment are not the same document with a different cover page, because the hazard is a moving mass close to people rather than a fixed guarded machine.

A cobot safety assessment for a stationary arm behind a light curtain answers a narrower question than a humanoid walking a shared floor needs answered: where the machine can reach is fixed for a cobot and constantly changing for a platform that walks, which is exactly the case the A2 Ultra's own footprint makes concrete.

Starting the concept first costs less because every one of these decisions is cheaper on paper than against a poured floor.

Moving a defined stop boundary two metres before the layout is drawn is an edit to a document. Moving it after conduit is run and a light curtain is mounted is a second construction project layered on the first one, and it is the reason a technically finished deployment sits idle waiting for a sign-off nobody scoped time for.

The regulatory deadline makes late timing worse than an internal scheduling problem.

The Machinery Regulation becomes mandatory on 20 January 2027, and there is no harmonised European standard yet for a free-walking general-purpose humanoid, which means the safety concept a project writes now is standing in for a standard that does not exist rather than filling in a template that does. A concept started early has time to be checked against that gap; one started after the build is a compliance question asked with the clock already running.

The safety concept is not a step that happens after the engineering proves the task is possible. It is the boundary the engineering has to be drawn inside from the first layout sketch, and a humanoid's own height and weight are exactly the numbers that boundary starts from.