One robot to start.
Then every robot.
The system is multi-robot from day one. New robots show up in the dropdown the moment a yaml lands — no code release, no app update. Every category — mass-produced, DIY, or private — uses the same fork tree, the same calibration walkthrough, the same skill format.
SO-101 today.
myCobot, Agility, yours next.
The content roadmap starts with SO-101 because that's the community we know. But the architecture is robot-agnostic. The dropdown is just a yaml directory.
Mass-produced and DIY-shared robots ride the open-data flywheel for free. Have a private one-off you don't want shared? The paid tier exists for that.
Your demonstrations make everyone's robot smarter.
Mass-produced and DIY hardware ride a mandatory-open data policy. AXION takes the public corpus, runs a full-parameter base reflux every 30 days or 3000 episodes, and re-publishes a stronger base. Your next skill starts on a better foundation than your last one.
Public-by-default episodes
Mass / DIY users contribute their episodes. Private (paid) tier keeps data sealed if you need it.
Admin curation + auto-anomaly
Operator queue + automated outlier checks pick the corpus for the next reflux training.
Full-parameter base FT
Managed cloud GPU node, ~8h per cycle. Per robot × backbone × camera-set.
axion-base-…-vN
New base auto-publishes to the marketplace. Existing skill forks re-attach via the fork tree.
*mandatory open applies to mass-produced + DIY-shared categories. Private hardware is exempt under the paid tier.