Dream Labs secures $90M for AI that turns robots into quick learners

Dream Labs, a startup building an AI foundation model for robotic control, has raised $90 million in seed funding. The round was backed by multiple investors including B Capital and Kindred Ventures. The company simultaneously launched its Dream API, offering developers access to its model at an hourly rate.

The funding represents significant early-stage investment in a company attempting to solve a key limitation of industrial robots: their inability to perform multiple tasks or transfer learned skills. According to reporting from SiliconANGLE, most of the estimated 5 million industrial robots worldwide are designed for single functions and require extensive programming to deploy. Dream Labs' approach aims to change this by enabling robots to learn from video demonstrations rather than specialized coding.

The company's model, Dream-1, analyzes videos of humans performing physical tasks to predict logical next steps and translate them into robotic actions. This method allows the AI to work with various two-armed robots using only video input and simple instructions. The company states this approach can significantly reduce deployment time for new robotic tasks.

Through the newly launched Dream API, developers can access different versions of the model at varying price points. For tasks not covered by the base model, the company offers a service where developers can submit video footage of new actions to create customized versions. To demonstrate the technology, Dream Labs has created a browser-based interface where users can test commands with a live robot.

The startup is collaborating with several robotic hardware manufacturers whose products fall within a specific price range. While the model is designed to work with different types of robots, the company works with partners to optimize performance for their particular machines. CEO Joel Jang has indicated the current model performs well at certain manipulation tasks, with plans to expand its capabilities to include different types of robotic movement and more complex hand designs.

The funding will primarily support the computational requirements of scaling the model. Investor Daisy Cai of B Capital noted that while robotic hardware has become more affordable, the intelligence to make them versatile remains a key need. The founding team's background includes work on robot learning at Nvidia, suggesting experience with similar technical challenges.

The size of this seed round appears notable in the robotics space, where funding amounts are often smaller even at later stages. Recent funding rounds for companies working on related technologies have varied in size and disclosure practices. Dream Labs' approach differs from traditional robotics companies by focusing on the AI layer rather than hardware development.

The practical adoption of this technology may depend on several factors. The pricing structure, while potentially competitive for certain applications, could present challenges for others. The company's progress in working with hardware partners and refining its model for different use cases might influence whether it can expand beyond its current focus areas.

For now, Dream Labs' funding and product launch indicate growing interest in AI solutions that could make robots more adaptable and easier to deploy across various industries.

Sources: siliconangle.com