NVIDIA Former Team Launches Dream-1 Embodied Model, Achieving Zero-Shot Task Execution Across Robots
英伟达世界模型团队创业！种子轮拿下9000万美元，具身模型反击通用模型
Dream Labs, founded by former NVIDIA DreamZero team members, has released the embodied model Dream-1, which supports zero-shot task execution across different robots. The model demonstrates strong generalization capabilities in both simulated and real-world environments, achieving the top score in the RoboDojo-RC Tier 1 test. Dream-1 learns to predict visual scenes and actions through a world action model (WAM), allowing it to perform tasks on previously unseen dual-arm robots without additional adaptation.
The release of Dream-1 highlights advancements in world models for robotics, enabling cross-robot generalization without prior data. The model's performance in real-world tasks and simulations suggests progress in sim-to-real transfer, though its effectiveness still depends on the robot's inverse kinematics solver.
Source: QbitAI — Robotics on WeChat · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
training data duration
30 minutes
View original evidence
今年2月发布的DreamZero，已经展示了这条路线的潜力：只用30分钟的YAM机器人数据做后训练，就能在这台机器人上尝试未见过的任务。
Open source S12model running cost
3
model running cost
2
simulation tests
27
real robot demonstrations
21
model training data duration
30 minutes
View original evidence
今年2月发布的DreamZero，已经展示了这条路线的潜力：只用30分钟的YAM机器人数据做后训练，就能在这台机器人上尝试未见过的任务。
Open source S12model training data duration
5 minutes
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
今年2月发布的DreamZero，已经展示了这条路线的潜力：只用30分钟的YAM机器人数据做后训练，就能在这台机器人上尝试未见过的任务。
Open source S12
Dream-1的模型运行费用为 每小时3美元 ，更轻量的Dream-1-flash则为每小时2美元。
Open source S10
在官网的跨机器人展示中，团队给出了27种未见过机器人的仿真测试，以及21段实机演示。
Open source S5
此外，在第三方评测机构Robocurve的RoboDojo-RC Tier 1实测中，Dream-1的平均任务得分为 59.3% ，在当前公布的模型配置中排名第一，超过大模型御三家的表现。
Open source S7
对于接下来的动作，他表示团队将继续缩小零样本使用与少量数据适配之间的差距，最终的目标，可能是连那5分钟的适配数据也不需要。
Open source S24
Source：QbitAI — Robotics on WeChat · mp.weixin.qq.com