HumanX combines XGen, a pipeline for producing physically plausible humanoid-object interaction data, with XMimic, an imitation-learning framework. Its research paper predates the October roundup.
The released repository documents examples including basketball catch-and-throw, a pump-fake shot, football kicking and bin handling in simulation. Building a policy from a new video also requires motion recovery, retargeting and training.
The research page reports transfer experiments on a Unitree G1 robot. That is not equivalent to a consumer robot immediately learning any sport by watching one clip. This overview preserves the distinction between released simulation workflows and demonstrated hardware experiments.
Sources: Author project page · Released implementation