There’s a weird problem that shows up when you crowdsource robot training: your hardest data probably shouldn’t come from your average contributor. @axisrobotics recently introduced something called the Challenger Program, and I think the idea behind it is more important than the name. Imagine two robot tasks. Task A is simple: pick up an object and place it in a bin. Task B involves awkward geometry, tight tolerances, multiple steps and a recovery if the first grasp goes wrong. If both tasks are thrown into the same public pool, you’re treating them as if they require the same operator skill. They don’t. Beginners hit the hard task, struggle with it, generate messy attempts and probably have a bad experience themselves. Experienced contributors, meanwhile, aren’t being specifically routed toward the tasks where their skill is most valuable. Axis is starting to separate those two worlds. Straightforward tasks stay in the regular pool. The difficult ones move into a dedicated Challenger zone for contributors who already know how to use the platform. At first this sounds like a small UX update. I think it’s actually pointing at something bigger: robot data has a skill-matching problem. Once a contributor network gets large enough, the question isn’t just: “Who can generate this data?” It’s: “Who should generate this specific data?” Easy tasks can optimize for scale. Hard tasks can optimize for operator skill. And eventually I can imagine the matching becoming much more granular: this contributor is great at precision grasping another is better at long-horizon tasks another is unusually good at recovery another performs well with a specific embodiment Then the network starts looking less like a crowd… and more like a distributed workforce with specialized skills. That’s a much more interesting way to scale human intelligence for robots. https://t.co/4FLsS5npxd