JoyIn Introduces Aether With a Partial Open-Source Plan

Outside developers may get access to parts of Aether built around touch, but not to its energy-use portions. JoyIn set that boundary when it publicly introduced the robotics model on Sept. 10, framing Aether as a perceptive system for humanoid control.

Touch access comes with a clear limit

JoyIn plans to open-source parts of Aether related to touch while keeping portions related to energy use out of that plan. The result is not an open release of the full system; it is access to a defined set of components. Whether those shared pieces can support useful work on their own will determine the practical value of the release for outside developers.

Aether’s pitch is physical control

JoyIn contrasts Aether with text-based robotic control, describing its approach as perceptive. The company says that approach helps humanoid robots complete tasks on their first attempt, rather than treating the model as a system for generating text. Aether’s development lead, Zhu Mingxuan, also said the model shortened training time by two-thirds.

The performance case remains JoyIn’s

The two figures describe different outcomes: one measures task completion on an initial try, while the other concerns time spent training. Both are company-reported results. The announcement does not turn either metric into an independently verified result, but it makes them central to JoyIn’s case for a perceptive-control model.

Two tests now follow the launch

Aether’s next test is twofold. JoyIn’s performance claims will need to stand on their own merits, while its partial open-source plan will show whether touch-related components are useful without the retained energy-use portions. The launch offers developers a potential building block, not access to every part of the model.