Robotics and automation: mechanics, control, vision, mobile fleets and the humanoid bet

A modern six-axis arm is repeatable to hundredths of a millimetre, runs for years, and costs a fraction of what it did two decades ago. Despite that, most manual work has not been automated, and the reason is consistent across industries: the arm is generic and everything else is bespoke. Fixtures, tooling, guarding, sensing and programming are engineered per application, and that cost does not fall as hardware does.

This branch is split to make that visible. Mechanics covers the components that decide precision and cost, and includes the most concentrated supplier market in the map — precision reducers, historically supplied by two Japanese firms. Control covers the servo layer and, more importantly, the programming problem above it. Vision covers the sensing that lets a cell tolerate a world that is not perfectly arranged. Mobile robotics covers the machines that move themselves rather than a part, which is a different problem again: a cell fails when it is misprogrammed, a fleet fails when four hundred vehicles want the same corridor.

The fifth segment is a different kind of thing. Humanoids and embodied AI are a bet that learned policies can remove the fixturing and programming rather than execute them faster. The hardware is demanding and tractable; the open question is whether a policy can be reliable enough for production and specified by demonstration rather than by an integrator. If it can, the economics of an enormous amount of manual work change.

The supply chain here is more Japanese and Chinese than anywhere else in this map. Reducers, servo motors, encoders and industrial sensors are dominated by a handful of Japanese firms, and the emergence of domestic Chinese alternatives over the past decade is the main structural change in the industry — and the reason humanoid cost projections look the way they do.