Anvil Bets on Nonhumanoid Robots
Mike Xia, CEO of Anvil Robotics, on humanoid vs. non-humanoid robots
Anvil is betting that the first big physical AI market will be built around fixed workstations, not full body robots. In food prep, light assembly, packing, and materials handling, the hard part is usually seeing, grasping, aligning, and placing objects, not walking around like a person. If most early deployments happen at benches, conveyors, carts, and packing stations, then extra legs add cost, joints, balance problems, and failure points without creating much more revenue.
-
The constraint is the job shape. Anvil keeps pointing to tasks where a worker stands or sits in one place, plugs in parts, opens bags, packs boxes, or portions food. Those workflows need manipulation, force sensing, and repeatability. They do not inherently need bipedal mobility, so Anvil focuses on the arm, sensors, controls, and developer stack around that narrower problem.
-
Humanoid builders are solving a different bottleneck. Figure, Apptronik, Agility, and Foundation sell the idea that a two arm, two leg robot can drop into human built factories with fewer retrofits, especially for moving through aisles, carrying items, and working across multiple stations. That makes sense when mobility is the scarce capability. Anvil is implicitly saying mobility is not the scarce capability in much of the near term market.
-
This also fits Anvil's business model. It sells arms and kits to small robotics teams that need to start collecting data fast. A legless manipulation platform is cheaper, easier to assemble, and easier to deploy in multiples, which matters when customers are buying several units for training, teleop, and proofs of concept rather than one expensive flagship robot.
If early physical AI adoption clusters around structured work cells, non humanoid systems should capture a large share of deployments before general purpose humanoids mature. That would make the winning stack less about building a single iconic robot, and more about supplying the cheapest reliable body, senses, and controls for hundreds of narrow, high frequency manipulation jobs.