Unitree G1-D puts humanoid on wheels

This is the more practical humanoid strategy: keep the arms, sensors, and training stack, but stop spending every watt on staying upright.

• –Wheels make G1-D less impressive in demo clips but much more useful for labs, warehouses, and embodied-AI data collection
• –Unitree is positioning the robot as part of a full data-and-training platform, not just another hardware showcase
• –The 3kg-per-arm payload, optional dexterous hands, and 0-2m workspace put it closer to mobile manipulation infrastructure than consumer robotics
• –The tradeoff is obvious: it loses stairs and rough terrain, but gains runtime, stability, and repeatable indoor navigation
• –For developers, the interesting bit is the tooling stack around robot datasets, simulation, distributed training, and deployment