Dream Labs raises $90 million to build a universal AI brain for robots

Breaking the programming barrier

Most industrial robots today are designed to perform one specific task perfectly. While there are roughly 5 million industrial robots running in factories worldwide, nearly every one is programmed for a single motion that cannot be transferred to another machine. This lack of flexibility means that teaching a robot a new task often requires specialists to spend months on manual programming.

Dream Labs is attempting to solve this by creating a universal AI brain for robots. Their goal is to make robots as easy to use as a rental car, where a driver does not need to relearn how to operate the vehicle just because the model changed. Instead of hard-coding every movement, they want robots to rely on general experience that can be applied to new hardware instantly.

The company recently secured $90 million in seed funding to scale this vision. The funding round was led by B Capital and Kindred Ventures, with participation from several other firms including A.Capital and Unusual Ventures. This capital will primarily be used for compute power to expand the variety of tasks and robots the system can handle.

How Dream-1 learns from humans

The core technology is a model called Dream-1. Unlike previous systems that might require training on a specific machine, Dream-1 was trained on video footage of people performing everyday tasks. By watching humans, the AI learns to predict what happens next in a sequence and determines the necessary physical actions to achieve a goal.

Because the training data comes from video rather than a single robot's sensors, the capabilities of the model are transferable across different brands and models. This allows for what the company calls zero-shot execution, where a robot can perform a task it has never seen before without any additional training. Out of the box, Dream-1 can control any robot equipped with grippers.

The model currently handles tabletop tasks like folding laundry or picking and placing objects. It is also being refined to support wheeled mobile robots and dexterous hands. For the 20 hardware partners the company already works with, the model is further tuned to ensure it performs at its best on that specific hardware.

Public access and developer tools

Dream Labs has officially launched the Dream API, which is the first publicly available tool of its kind. This API takes a robot's camera feed and a plain-English instruction to generate motor commands. Developers can sign up today to connect their own robots with grippers to the system to see them carry out tasks.

To support hardware that is not yet officially integrated, the company offers a Fine-tuning as a Service. This allows robot makers to submit a small batch of demonstration data to receive a tailored model within a few hours. This service helps bridge the gap for specialized hardware or unique capabilities that the base model might not cover perfectly.

The company also provides a Robot Playground for those who do not own any hardware. Users can access a browser-based interface to type commands to a robot located in the company's San Francisco office. This allows people to watch the robot respond in real-time, providing an unedited look at how the AI handles various instructions and potential mistakes.

Impact on the robotics market

The cost of capable robot arms has dropped significantly, moving from the price of a luxury car to the price of a high-end laptop. However, the hardware itself remains

Scaling the robot brain

The company intends to use its new funding to expand the scope of what Dream-1 can do. While it currently handles basic tabletop tasks, the goal is to scale the system to support a much wider variety of users and complex environments. By providing a working model that people can test themselves, they aim to move past edited demos and into real-world utility.

Early adopters are already reporting that the technology significantly reduces setup time. One hardware partner noted that their robots were up and running in a few minutes, attempting tasks that had never been programmed before. This shift allows hardware manufacturers to focus on building better physical machines while Dream Labs provides the intelligence that makes them useful.

As the company grows, it will continue to work with over 20 hardware partners to refine the model. The focus remains on making robots that cost between $2,000 and $10,000 capable of doing useful work in minutes rather than months. By making the AI brain universal, they hope to create a standard where any robot can be