Dream Labs Unveils Model for Controlling Unfamiliar Robots

Startup Dream Labs, founded by four former NVIDIA researchers from the GR00T team, has emerged from stealth with a $90 million seed round and launched a public API for the Dream-1 model, which controls bimanual robots in zero-shot on machines it has never been trained on.

Compiled by Sergey KostenchukPublished 2026-10-21Updated 2026-10-21

2026-10-21 Research Google

🤖 Dream Labs Unveils Model for Controlling Unfamiliar Robots

A startup founded by four former NVIDIA researchers from the GR00T team emerged from stealth on October 19, 2026, with a $90 million round and launched a public API for the Dream-1 model: it learns from video and controls bimanual robots in zero-shot on machines it has never been trained on.

🌍 Zero-shot via a cloud API breaks the robotics "chicken and egg" problem: an affordable bimanual machine (arm for ~$2,000) is useful from day one. On the independent Robocurve benchmark, Dream-1 scored 59.3% compared to ~48% for Gemini 4 Argon, with an attempt costing ~$0.15.

👤 Owners of a bimanual robot with parallel grippers and an IK solver can connect it to Dream-1 for $3/hour without fine-tuning — the API is available at platform.dreamlabs.ai. The model is marked as research beta: it has been tested on real machines on only 4 robots.

Source 1: https://dreamlabs.ai/research/dream-1/

Source 2: https://robocurve.org/

Sources