VLA model (vision-language-action)

Instead of programming each step, you tell the robot something like put the red cup in the sink, and the model decides how to move the joints. The same model can handle many tasks, which is why VLA models are at the center of current humanoid projects.

For a buyer this is still early. Demos often work in the setting where the model was trained and fail in a new kitchen. Ask for success rates on tasks in unfamiliar places, how long one action takes, and whether the model runs on the robot or needs a cloud connection.