Dream Labs
Dream Labs is building a general-purpose robot brain to make robotics more accessible and useful in everyday life. Its World Action Models translate camera observations and plain-English instructions into robot actions across different robot embodiments.
At a Glance
- Consumers and prosumers buying affordable robots
- Small robotics teams and startups
- Robot manufacturers and hardware builders
- Robotics researchers and university labs
- +2 more
AI Tools by Dream Labs
(1)
Dream-1
Robot Foundation Model API
Discussions
No discussions yet
Be the first to start a discussion about Dream Labs
Latest News
Dream Labs raises $90M seed to build the brain for everyday robots and launches Dream API
Dream Labs comes out of stealth and introduces Dream-1, a public World Action Model API for unseen robots and tasks
Kindred Ventures announces its investment in Dream Labs' $90M seed round and Steve Jang joining the board
The founding team releases DreamZero, the research paper introducing World Action Models
Products & Services
Dream Labs' first World Action Model available through a public API. It uses camera images, robot state and a task instruction to return actions, and is designed to generalize zero-shot to unseen environments, tasks and bimanual parallel-gripper robots.
A public developer API for connecting robots to Dream-1; the Python client sends observations and receives robot actions without requiring machine-learning expertise.
A fine-tuning service where customers upload demonstrations for an existing or new robot embodiment and receive a fine-tuned model ID.
An interactive web playground for prompting a real robot without the user needing to own one.
Market Position
Dream Labs positions Dream-1 as a hardware-agnostic, general-purpose robot foundation model: a shared brain that can control robots from different manufacturers and generalize zero-shot, rather than a vertically integrated robot maker. Its closest competitive set includes robot-foundation-model companies such as Skild AI and Physical Intelligence; Dream Labs differentiates around public API access, low-cost hardware, rapid deployment and zero-shot embodiment transfer.
Leadership
Founders
Joel Jang
Co-founder and CEO; formerly a Senior Research Scientist at NVIDIA's GEAR Lab, where he led the world-model team for Project GR00T. He led work on DreamZero, DreamDojo, DreamGen and LAPA.
Scott Reed
Co-founder, President and Chief Scientist; formerly Principal Research Scientist/Research Scientist at NVIDIA, where he worked on machine learning, unsupervised feature learning and deep learning and contributed to the GEAR/GR00T research program.
You Liang Tan
Co-founder; an AI and robotics researcher who previously worked with NVIDIA's GEAR research group, and whose background includes research at UC Berkeley and graduate study in computer science at Georgia Tech.
Yunhao (Andy) Ge
Co-founder; previously a Research Scientist at NVIDIA, working on NVIDIA Cosmos and Deep Imagination Research, with prior research/internship experience at Google Research and Google Cloud AI.
Executive Team
Joel Jang
Co-founder and CEO
Former NVIDIA GEAR Lab Senior Research Scientist and world-model lead for Project GR00T; led DreamZero and related world-model research.
Scott Reed
President and Chief Scientist
Former NVIDIA Principal Research Scientist/Research Scientist focused on machine learning, deep learning and robotics/world-model research.
Board of Directors
Founding Story
The founders came from NVIDIA's GEAR Lab, where they developed DreamZero, a 14-billion-parameter World Action Model. After seeing DreamZero adapt to a new YAM robot using only about 30 minutes of robot data, they concluded that generalizable models plus increasingly affordable robot hardware could break robotics' chicken-and-egg problem; after six months of building, they launched Dream Labs and Dream-1.
Business Model
Revenue Model
Usage-based API access for robot inference, complemented by paid fine-tuning services for customer robot embodiments. Dream Labs makes Dream-1 available through a public developer API and positions the platform for robot builders and operators.
Pricing Tiers
Published price for the flagship Dream API model.
Target Markets
- Consumers and prosumers buying affordable robots
- Small robotics teams and startups
- Robot manufacturers and hardware builders
- Robotics researchers and university labs
- Developers building robot applications
- Enterprises evaluating general-purpose physical AI
- Everyday household tasks such as folding clothes, clearing dishes, wiping counters and restocking shelves
- Robot learning and research on unseen tasks and embodiments
- Developer and startup prototyping with affordable bimanual robots
- Deploying a shared robot brain across hardware from multiple manufacturers
- Fine-tuning a general model for a customer's new robot or task
- Hugging Face
- Scale AI
- KAIST
- Anvil Robotics