Rhoda AI raises $450M to construct foundational robotics models that learn from web videos

Palo Alto-based Rhoda AI, a developer of foundational artificial intelligence robotics models, said today it raid $450 million in Series A funding to coach intelligent robots using publicly available web videos.

Premji Invest led the round, bringing the valuation of the startup to $1.7 billion.

The web has been a prolific ground for providing examples of human activity. It’s an ideal place to get candid presentations of virtually every task, from picking up items of all types to packing boxes, sorting objects and completing various manual jobs.

Currently, most robots are trained through a process called teleoperation. This requires specialized remote-control equipment, including gloves and external sensors that collect positional data that will be translated into movements. Some systems also can work from quite a few videos of actors taking the identical motion again and again.

This limits the full amount of coaching data that will be input into robotics models. To handle this, most robotics corporations extrapolate from these limited data sets by generating synthetic datasets.

Nevertheless, with the vast amount of information already available on the web of a whole bunch of 1000’s to hundreds of thousands of humans already doing household tasks in natural environments, researchers have ready-made datasets.

Rhoda’s team said they imagine that online videos will bridge the gap between limited data for orienting tasks by supplementing the small amount of robot telemetry with a broader array of natural human movements.

“Within the case of teleoperation, if the phone orientation changes, that is likely to be enough to cause the model to fail,” Chief Executive Jagdeep Singh told Bloomberg. “Whereas in our case, the model it’s seen so many other examples of objects which might be at different orientations, it’s capable of generalize.”

Teleoperation may give a robot a framework to grasp the fundamentals of a task, nevertheless it cannot give a robot the wherewithal to adapt. That’s where online videos are available in to offer the context, generalization, different orientations, failure states, varied approaches and edge cases. Something that’s difficult to breed even with the “fuzzy logic” of synthetic generation.

The corporate calls the approach Direct Video Motion and said it allows setups similar to robot arms to cope with edge conditions higher than current AI training models. The corporate added that it has run successful tests using off-the-shelf parts with an automotive firm and has plans to license its AI model to customers.

Singh said that Rhoda plans to construct humanoid-style robots and has plans to make its own hardware to make sure quality for real-world work.

Other backers joining the round included Khosla Ventures, Singapore’s state-owned investor Temasek Holdings Pte and enterprise capitalist John Doerr.

Rhoda joins a variety of corporations drawing investor attention for a growing advancement called Physical AI, where AI bridges the gap between the physical and digital worlds, allowing machines to perceive, reason and interact with 3D environments in real-time. Although reasoning and perceiving robots first come to mind, physical AI includes self-driving cars, adaptive assembly lines, robotic surgery, and smart buildings. Where traditional AI “thinks,” Physical AI is able to “considering, then acting.”

Image: SiliconANGLE/Microsoft Designer