Robot skills vs robot apps: what an โapp storeโ would really sell
In robotics, an โappโ is rarely a standalone program like on a phone. Itโs closer to a behavior module that plugs into a larger stack: perception, navigation, planning, safety, and hardware control.
A better mental model than โappsโ is skills. A skill is a parameterized behavior that runs for some time, gives feedback while it runs, and can be stopped safely. That maps closely to how robotics middleware already thinks about โdo a thingโ behaviors
So when people say โrobot app store,โ the most realistic first step isnโt a library of full applications - itโs a library of installable skills. You can already see early versions of this idea:
โค General robot marketplaces have existed for years (e.g., RobotAppStore) as distribution for robot software.
โค In late 2025, Unitree launched a public beta โApp Storeโ focused on downloadable action routines / skills for its humanoid robots, which is closer to โskill distributionโ than mobile apps
โAppโ then, is usually just a bundle: a set of skills configuration UI integrations (doors/elevators/fleet tools) packaged together. The reason this matters is that skills are only portable if they come with a clear contract, something like:
โค what inputs they expect (camera, map, arm pose),
โค what parameters they accept (โtarget=blue_bottleโ, โspeed=lowโ),
โค what feedback they expose (progress, confidence),
โค what โsafe stopโ means for this skill.
Marketplaces already exist that advertise โrobot apps/behaviorsโ broadly, but the practical unit that scales across robots is still the skill - because skills are the smallest thing you can install, run, monitor, and cancel without turning deployment into a custom project.