The robots most companies show you work in a lab. Rhoda is building the ones that work where the boxes are crooked, the light is wrong, and nothing sits still.
RHODA AI. Two robot arms, a cardboard box, and a model that learned how the world moves by watching millions of videos. The whole pitch, in a single frame.
On a factory floor somewhere in California, a pair of robot arms reaches for a cardboard box. The box is slightly out of place. The lighting has shifted since lunch. A human worker would not notice either thing - they would just adjust. For most robots, that small disorder is the end of the demo. For Rhoda's, it is just Tuesday.
That is the quiet, unglamorous problem Rhoda AI has decided to make its life's work: not the dancing humanoid that goes viral, but the dull, mission-critical motion that actually keeps manufacturing and logistics running. The company emerged from 18 months of stealth in March 2026 with a $450 million Series A, a $1.7 billion valuation, and a thesis that sounds almost too simple - teach robots the way the internet taught everything else. By watching.
We believe the next era of robotics requires models that understand how the world moves - not just what it looks like or how it's described in language.
Most robots memorize. You show them a task, over and over, in a tidy lab. They get very good at that exact task and fall apart the moment reality wanders off-script.
Rhoda's bet is the opposite. Its system, FutureVision, is pre-trained on hundreds of millions of internet videos - so before it ever touches a real arm, it already has a rough physics of how objects fall, slide, bend, and bump. Rhoda calls the architecture a Direct Video Action (DVA) model: it watches a scene, predicts what happens next, and turns that prediction straight into motion. Then it does it again. And again. Every few hundred milliseconds, in a closed loop, the robot revises its next move as the world changes.
It is less "follow these instructions" and more "keep imagining the next second and act on it." The contrarian part isn't the ambition - it's the target. While the industry chases telegenic humanoids, Rhoda points its model at the work nobody films: the component-processing line, the box that has to move now.
Cameras feed a live view of the scene into the model.
FutureVision predicts the next states of the world on video.
DVA converts those predictions directly into robot actions.
The loop refreshes every few hundred ms as conditions shift.
Investor Mayfield framed the prize bluntly: the global market for manual labor runs around $30 trillion a year, with more than $10 trillion of it in the United States alone. You don't need to win much of that to matter.
The deployment math is just as direct. By Mayfield's estimate, roughly 1,000 deployed units would imply about $100M in recurring revenue; scale to 10,000 and you're looking at $1B+ ARR. Rhoda's longer game is to stop being only a robot company and start licensing its intelligence layer to other people's hardware.
A foundation model built on video-predictive control. It carries long-context visual memory, predicts where a scene is heading, and runs the robot in a closed loop - revising actions every few hundred milliseconds. It powers Rhoda's own systems today and is meant to be licensed across other platforms over time.
The proprietary model that links perception straight to control. Pre-trained on internet-scale video for physics and motion priors, it converts predictions into real, physics-aware movement - and can pick up a new task from as little as ~10 hours of teleoperation data.
CEO Jagdeep Singh is a serial deep-tech founder - best known for solid-state battery company QuantumScape - now pointing his track record at robot brains. Alongside him: the computer-vision research bench that makes the video approach plausible.
A $450M Series A, led by Premji Invest, with a roster that reads like a deep-tech who's-who.
Rhoda AI Corporation is founded and goes heads-down in stealth.
18 months building FutureVision and the Direct Video Action model; quiet pilots in live manufacturing environments.
Exits stealth with a $450M Series A at a $1.7B valuation, led by Premji Invest. FutureVision unveiled publicly.
A component-processing workflow completed in under two minutes per cycle, autonomously - reportedly exceeding customer KPIs.
The founder who once chased a better battery is now chasing a better robot. From QuantumScape to robot foundation models - across, by one account, roughly seven companies.
Instead of staged demos, Rhoda pre-trains on hundreds of millions of internet videos. The robot effectively learns physics by binge-watching.
The model "thinks" in a loop, rewriting its next move every few hundred milliseconds - closer to reflex than to a script.
The pitch is deliberately boring on purpose: skip the viral humanoids, automate the unglamorous work nobody films.
Return to that factory floor. The box is still slightly out of place. The light has still shifted.
The difference is that the disorder no longer ends the demo - it is the demo. Rhoda's arms reach, the box is handled, the cycle closes in under two minutes, and nobody calls a technician. If the company is right, the most important robots of the next decade won't be the ones doing backflips on a stage. They'll be the ones quietly handling the crooked box, in the wrong light, while the rest of us aren't watching - because they already watched everything else.