Robotics Consolidates Around Tactile Hardware
Mike Xia, CEO of Anvil Robotics, on humanoid vs. non-humanoid robots
This is the point where robotics stops being a science project about every possible sensor, and starts looking like a scaled supply chain for the few inputs that actually move task success to production grade reliability. Anvil’s argument is that once force or tactile prove they raise success from decent demos to near perfect execution, the winning companies will not just have better models, they will have cheaper, easier to deploy sensor stacks and enough hardware in the field to generate large volumes of that data.
-
Today’s vision first bias is largely an economics story. Cameras are cheap and easy to integrate, while force torque sensors still cost thousands of dollars and are hard for small teams to deploy across dozens or hundreds of robots. That keeps the industry in an experimentation phase instead of a scaling phase.
-
The same pattern showed up in LLMs and is now showing up in robotics. Once a data recipe proves it consistently improves outcomes, the market standardizes around it, vendors optimize cost and reliability, and application companies build on top instead of reinventing the stack for every deployment.
-
This also explains why Anvil sits below the application layer. Many robotics companies need working bodies, sensors, and controls so they can focus on customer workflows like packing, assembly, or materials handling. The moat shifts toward owning repeatable hardware and the data collection loop around real deployments.
Over the next few years, physical AI should consolidate around a narrower set of sensing stacks, and that will make the market look more industrial and less exploratory. The companies that win will be the ones that can mass produce the right sensorized hardware, get it into real workflows quickly, and turn every deployment into more training data.