Dream Labs raises $90 million seed and launches Dream-1 robot AI API

Dream-1, described as a World Action Model, predicts video and robot actions together. It uses three camera views to infer end‑effector poses—position, orientation and gripper state—so the same output can be mapped onto different robot arms via each machine’s inverse‑kinematics solver. The current zero‑shot capability targets bimanual robots with parallel grippers; tests on 27 unseen robot embodiments in simulation ran 300 rollouts each, showing varied performance across hardware. Developers can also use a fine‑tuning‑as‑a‑service, uploading a few minutes of robot‑specific data to adapt the model.

Inference is priced at $3 per hour for Dream-1 and $2 per hour for the lighter Dream-1‑flash, with each call returning 1.6 seconds of actions. Dream-1 remains a research beta and has not been validated for safety‑critical use, but the funding gives Dream Labs resources to broaden support beyond two‑arm, parallel‑gripper systems.