Democratizing Force and Tactile Sensing
Mike Xia, CEO of Anvil Robotics, on humanoid vs. non-humanoid robots
This reveals that force sensing still lives in the old automation world, not the new developer tooling world. In practice, many force torque sensors are sold as part of a custom factory cell, where an integrator installs the robot, wiring, controls, and safety systems for one exact station, which is why adding force data has historically meant enterprise style budgets, long setup cycles, and specialist engineering rather than something a small robotics team can buy and use in a week.
-
Anvil is selling into teams that want to skip that custom integration step. Its core pitch is that robotics teams otherwise spend five to six months and several engineers piecing together cameras, arms, compute, firmware, and controls before they can even start collecting data and training models.
-
That makes force sensing strategically important but commercially awkward. Research shows force data improves task success and helps models avoid failure modes, but the hardware has usually been too expensive and too hard to integrate to become a default part of the physical AI stack.
-
The contrast with humanoid companies is useful. Figure, Apptronik, and Agility are trying to drop robots into existing labor workflows, while component suppliers like Anvil are abstracting away the messy low level hardware choices so solution companies can build a customer specific robot system on top.
The next phase is turning force and tactile sensing from custom industrial equipment into standard developer infrastructure. As that happens, more robotics teams will train on richer real world data, and the advantage will shift toward the platforms that make those sensors cheap, reliable, and easy to deploy across many robots instead of one factory station at a time.