Foot Contact Sensors

This component does not bind, and that is the finding. Cumulative production runs about 2.1× ahead of cumulative need even at 2040, there is no global crossover at any point in the horizon. Only the single-bloc views bite: building from the United States alone falls behind in 2040, and from China alone in 2048. It therefore belongs among the single-bloc chokepoints rather than the binding spine, and it is worth saying plainly that we went looking for a shortage here and did not find one. Two corrections shaped that result. First, the scope: our own component-gap sweep justified adding this part as “distinct from the six-axis arm force/torque sensor already modelled”, but the force/torque link in this report is not arm-only. It explicitly carries two wrist plus two ankle units per humanoid², so the ankle transducer that dominates humanoid foot sensing was already counted, and folding it in here would have double-counted the rigidest sensor link in the report. This section is therefore scoped strictly to the sole-level element beneath the ankle unit, the pressure matrix or multi-point contact module¹, which is genuinely unmodelled and genuinely distinct. Second, penetration: sole instrumentation is a design choice, not a given, and the highest-volume legged robots skip it entirely⁴. What is left is a real but modest draw against a printed-electronics supply base that is simply too large to strain.

A legged robot needs to know when its foot has landed, how hard, and whether it is slipping. There are three ways to sense that²: a six-axis force/torque transducer at the ankle, a planar pressure matrix of force-sensing resistors in the sole, or planar strain sensors. The first is a precision instrument, expensive, calibration-gated, and already modelled in this report as the force/torque link. This section is about the second. A documented sole design is a four-point foot module built on contact-resistance force sensors¹, giving four discrete sensing points per foot and about eight per fully-instrumented humanoid; the elements themselves are screen-printed on polymer film, closer in manufacture to printing than to machining. The two classes are complementary rather than competing: the ankle transducer measures the total load path through the leg, while the sole array resolves where on the foot that load sits, which is what zero-moment-point balance control and terrain estimation actually consume. On quadrupeds the picture is different again, ANYmal fuses a six-axis transducer built into the foot with inertial data⁴⁷, but that is the premium research configuration, not the volume one.

The per-unit counts here are penetration-weighted, and the penetration is the entire uncertainty on this link. On humanoids, a fully instrumented machine carries about eight sole elements¹, but many designs run the ankle six-axis alone and infer contact from it, so the fleet blend is held at about three per robot, roughly 40% penetration. On quadrupeds the case is starker: the unit-volume leader does not use foot force sensors at all. Unitree, roughly 70% of quadruped sales, detects contact on the Go1 and Go2 from leg-mounted inertial sensors and joint torque rather than any foot sensor, on the reasoning that an IMU captures six degrees of freedom where a foot force sensor typically measures only normal force⁴. Only premium platforms instrument the foot, so the quadruped blend is about 1.2 elements per unit. Across the consensus ramp⁸ that produces the fleet’s cumulative need for sole elements. The quadruped share of that is about 3%, this is the first component in the report where a non-humanoid embodiment carries a real sourced bill-of-materials entry for a new part, and it still does not change the shape of the curve. If the fleet converges on estimation rather than instrumentation this link shrinks toward irrelevance; if it converges the other way, on universal four-point soles, the count nearly triples. That is a design-convergence question, not a supply question, and no amount of supply-side research resolves it.

Production is the only thing that differs between the Global, China and US views. The worldwide fleet and the elements it needs are identical in all three. The force-sensing-resistor market runs about US$525 million a year, growing 10.3%⁵, which at an instrumentation-grade blended price implies on the order of 60 million relevant elements a year once the consumer-electronics slice, 42% of revenue, and a materially cheaper spec than a robot foot must survive, is excluded. That base grows to roughly 195 million a year by 2040 on the reported growth rate, and it is genuinely elastic: these are screen-printed parts, so adding capacity is a matter of presses and screens rather than long-lead precision machines. The supplier map is unusually healthy for the United States. The force-sensing resistor was invented and commercialised in America, by Interlink Electronics of Camarillo California, which still manufactures printed force sensors as one of its two divisions and is independently listed⁶; Tekscan of Boston is the other named leader, and Sensata is US-listed⁵. China holds roughly 35% of units through the same flexible-sensor cohort that supplies this report's tactile-sensor link, and Asia-Pacific leads on revenue share⁵. The premium legged-robot tier sits in Europe, with Bota Systems, an ETH Zurich spinout, supplying ANYmal⁷. Every unit figure here is derived from a revenue number, not reported, and the published market sizes for this category disagree by more than tenfold, so the level is soft even though the direction is not.

Globally, never. Cumulative production stays ahead of cumulative need through the entire horizon, meeting about 586% of the 2040 requirement. The single-bloc views are where the interest lies. US-only crosses in 2040, meeting 86% of the 2040 need, not because American capability is thin, which it is not, but because an 18% share of a global base cannot equip a worldwide fleet on its own. China-only crosses in 2048, the latest single-bloc crossover of any component in this report. What this means in practice is that foot sensing is a concentration question rather than a capacity question: there is no world in which the industry cannot make enough of these, only worlds in which one bloc chooses to make them all. That places this component alongside magnets, compute and lidar in the single-bloc tier, though at the mildest end of it, and it is a useful counterexample to the report's dominant pattern, showing that a genuinely novel, genuinely robot-specific part need not be a bottleneck.

This link is a consumable, and it has the best-evidenced replacement clock in the report, better than the tendon drives, whose clock rests on a cycle rating rather than field failures. A robot foot is the harshest duty station on the machine: foot-ground impacts routinely exceed the nominal ground-reaction force by three to five times, repeated exposure to which is documented to cause outright sensor failure, and strain-gauge elements are specifically called out as susceptible to impact damage³. The historical proof point is blunt: bipedal robots in the DARPA Robotics Challenge carried force/torque sensors in their feet and could not walk for extended periods because those sensors kept failing³. We model a three-year mean service life on that basis, shorter than the four years used for tactile sensors and tendons, because impact loading is more destructive than abrasion or flex. And it still does not bind. Even with the replacement flow compounding across the horizon, a printed-electronics base measured in the hundreds of millions of units a year simply absorbs it. That is the useful lesson of this section: a component can have a punishing failure mode, a documented history of breaking in the field, and a consumable re-buy clock, and still not be a supply-chain bottleneck, because the thing that is scarce here is not capacity but engineering. The real constraint is the durability-versus-sensitivity trade-off: the conventional fix for impact survival is to widen the sensing range, which costs exactly the sensitivity the sensor exists to provide³. Robot feet do not need more factories. They need a better sensor.