Frontier Labs
Rhoda AI Exits Stealth With $450M to Train Robots on Video
AI robotics startup Rhoda AI has emerged from stealth with $450 million in Series A funding at a $1.7 billion valuation, using a novel approach that trains robots by learning from millions of publicly available internet videos.
Rhoda AI Launches With $450M to Bring Robots Into the Real World
Rhoda AI has emerged from stealth with a $450 million Series A that values the Palo Alto-based robotics startup at $1.7 billion. The round was led by Premji Invest, with participation from Khosla Ventures, Singapore's state-owned investor Temasek Holdings, venture capitalist John Doerr, Mayfield, and Capricorn Investment Group.
The company is tackling one of the hardest problems in AI: getting intelligent robots out of controlled laboratory settings and into real-world industrial environments where conditions are unpredictable. CEO Jagdeep Singh believes the key lies in an unconventional training methodology that leverages the vast library of human activity footage already available on the internet.
Direct Video Action: Learning From Millions of Internet Videos
Rhoda's core innovation is its Direct Video Action (DVA) model, which trains robots using publicly available internet videos rather than traditional teleoperation or synthetic data. The approach works by having robots continuously observe their surroundings, forecast future physical states as video sequences, convert those predictions into actions, and execute them in a closed loop.
The advantage over traditional methods is scale and diversity. Teleoperation -- where a human manually guides a robot -- produces limited training datasets that struggle with edge cases. As Singh explained, with teleoperation "if the phone orientation changes, that might be enough to cause the model to fail," but Rhoda's video-based approach exposes models to millions of diverse real-world examples, enabling better generalization across different orientations and unexpected conditions.
Automotive Pilots and Humanoid Ambitions
Rhoda has already completed successful pilot testing with an automotive firm using off-the-shelf robotic components, demonstrating that its DVA models can transfer learned behaviors to physical hardware without custom robot designs. The company plans to eventually build proprietary humanoid-style robots and license its AI models to customers who want to add intelligence to their existing robotic platforms.
The funding will accelerate research and engineering, expand industrial deployments and customer pilots, and grow Rhoda's multidisciplinary team spanning generative AI, computer vision, and robotics. The $450 million raise makes it one of the largest Series A rounds in AI robotics history, reflecting investor enthusiasm for "physical AI" that moves beyond text and code into factories, warehouses, and field operations.
What This Means for Engineers and Job Seekers
Rhoda's emergence signals growing demand for engineers who bridge computer vision, video understanding, and robotic control systems. The company's DVA approach -- training on internet-scale video rather than curated datasets -- points to a new paradigm in robotics where skills in large-scale data processing, predictive modeling, and sim-to-real transfer are increasingly valuable. As physical AI investment accelerates, expect more roles combining ML research with hands-on hardware integration.