The robotics stack, in the open.

1.4B-parameter foundation model for manipulation across 700+ object categories.

Production-grade manipulation pipeline. Perception, grasp planning, motion planning, control.

Build embodied agents with VLA backbones, tool use, episodic memory, and long-horizon planning.

GPU-accelerated robot simulator with differentiable physics and photorealistic rendering.

Pretrained vision-language-action checkpoints with a unified inference API.

Distributed coordination primitives for fleets of heterogeneous robots.