Robot AI trained on millions of videos aims to work beyond labs

Robotics startup Rhoda AI has introduced FutureVision, a novel robot intelligence system designed to enable machines to operate reliably outside controlled laboratory environments. Unlike traditional industrial robots that follow pre-programmed paths and struggle with changing conditions, Rhoda’s system continuously observes its surroundings, predicts future physical states through video, and adapts its actions in real time. This closed-loop approach allows robots to maintain accuracy amid unexpected changes such as shifting layouts or irregular workflows. The company’s innovation lies in pre-training its model on hundreds of millions of internet videos to learn motion and physical interactions before fine-tuning with smaller amounts of robot-specific data, significantly reducing the need for extensive teleoperation training. Rhoda AI has demonstrated the effectiveness of FutureVision in production settings, where robots handled variable materials and workflows without human intervention, surpassing customer performance targets. The startup recently secured $450 million in Series A funding to scale its technology, expand industrial deployments, and develop FutureVision as a foundational model for licensing to hardware and software partners