Dream Labs Completes 90000000 USD a seed funding round and opens Dream-1 robot model API

Article

NVIDIA GR00T Researchers launch a startup: raises 90000000 USD, launches a cross-robot model AI Beating AI breaking news: NVIDIA GR00T project, founded by four former researchers,Dream Labs announced the completion of 90000000 USD a seed funding round and opened Dream-1's API.

Developers need only provide robot camera footage, its current state, and task instructions, and the model can generate robotic-arm actions. The standard version per hour 3 USD, and the lightweight version 2 USD.

Previously, switching to another robot often required recollecting data and adapting the model. This year,February, the DreamZero launched by NVIDIA, requires 30 minutes of data from the target robot for additional training.Dream-1 eliminates this step. It learns the relationship between video and actions through a world action model, predicts where the gripper should move, and then relies on the robot's own control program for execution.

The team tested 27 robots unseen during training in a simulated environment, including 2008 launched in PR2. Testing partners from 8 countries also demonstrated real-robot operations such as stacking cups, folding paper boxes, and pulling tissues. However, it currently mainly supports dual-arm robots equipped with parallel grippers.

In third-party organization Robocurve's real-robot evaluation,Dream-1's average task-progress score was 59.3%, exceeding Google's Gemini 4 Argon's 48.2%, while the estimated model cost per attempt was only 0.15 USD, approximately 1/16 of the latter.

However,Dream Labs separately disclosed 400 tests showing that the full success rate for pick-and-place tasks was only 40%, and as low as 6%when using tools.

The model remains in the research testing stage and has not yet been validated for safety-critical scenarios.

Original link https://m.theblockbeats.info/flash/372761

Key points

Dream Labs Founded by four former researchers from NVIDIA's GR00T project.
Dream Labs Announced the completion of 90000000 USD a seed funding round and opened Dream-1's API。
After developers provide robot camera footage, its current state, and task instructions,Dream-1 can generate robotic-arm actions.
Dream-1 The standard version per hour 3 USD, and the lightweight version per hour 2 USD。
The team said that Dream-1 does not require 30 minutes of data from the target robot for additional training.
The team tested 27 robots unseen during training in a simulated environment.
Robocurve's real-robot evaluation shows that Dream-1's average task-progress score was 59.3%，Gemini 4 Argon's was 48.2%。
Dream Labs The disclosed 400 tests show that the full success rate for pick-and-place tasks was 40%, while the full success rate when using tools was 6%。
Dream-1 It remains in the research testing stage and has not yet been validated for safety-critical scenarios.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

Dream-1 It receives footage, robot state, and task instructions, predicts where the gripper should move, and then the robot's own control program executes the action. The report emphasizes that it can be used across different robots, but it currently mainly supports dual-arm robots equipped with parallel grippers.

Why it matters to readers

If cross-robot capability holds up in more real-robot tests, developers may no longer need to conduct the reported additional training for the target robot when switching robots; however, the full success rate in disclosed tests remains limited.

Beginners should distinguish between “task-progress score” and “full success rate”: a high former score does not mean that a task can be completed in full, and the research testing stage likewise does not mean it is already suitable for safety-critical scenarios.

Risks and unknowns

The report did not disclose the list of seed-round investors.
The funding information, product capabilities, and evaluation figures are all relayed by the report; the original article did not include original documents from the company, investors, or evaluation organization.
27 The robot testing was conducted in a simulated environment; performance across different real-world robot models still requires further validation.
400 In this round of testing, the end-to-end success rate for grasping and placement was 40%, while the end-to-end success rate for tool use was 6%。

Dream Labs Completes 90000000 USD a seed funding round and opens Dream-1 robot model API

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center