It must use a good old computer + mouse + keyboard or some other setup a human would normally use. it must complete minecraft by itself with no external help during the successful run.
Update 2026-05-21 (PST) (AI summary of creator comment): The robot must control its actions through muscle-like actuators (not just software wrappers that bypass physical actuation). The model must physically operate a standard computer setup (mouse, keyboard, or similar human-used interface).
People are also trading
Are there any requirements for the software controlling the robot? In principle you could have lightweight models/wrappers that convert the input/output of the robot sitting at a computer to something a native computer use model can interact with. So that model gets a 2d image of the screen (extracted from robot's camera view) and can send command like "press W" that get translated into finger actuation. I think that would reduce a yes resolution to being only trivially harder than "having any computer use model that can complete minecraft" but not sure if it's in the spirit of the question.
@2b3o4o as long as the model is controlling the robot through activating muscle-like actuators, and it completes minecraft on a good old computer + mouse + keyboard, or some other setup a human would normally use, it counts