Giving AI a Body

A guided map of embodied AI today: the vision-language-action and world-action models that drive robots, the data and senses they learn from, and the agent harnesses and robot fleets that put them to work in the physical world.

Background: Vision-language-action models · World models

In order

1 stories

Embodied · Open VLA models

Ai2's MolmoAct2 is a fully open robot model aimed at low- to medium-cost hardware

A preprint pairs a spatially trained vision-language model with a motion generator and reports strong out-of-the-box results, though its real-world trials are small.

arXiv preprint · 4 May 2026