Rhoda Robotics Valued at $1.7B After Funding
AI robotics startup Rhoda, which trains robots using online videos, just hit a $1.7B valuation in its latest funding round. This underscores the growing interest in machine learning systems that can learn from real-world data for robotics applications. It’s a sign that investors are betting big on AI's ability to revolutionize robotics.
Rhoda AI, fresh out of stealth mode after 18 months, is pioneering a new approach to robotic intelligence called FutureVision. This system uses video-predictive control, enabling robots to operate effectively outside controlled lab environments. The company's architecture, known as a Direct Video Action (DVA) model, bridges perception and control by continuously observing its environment, predicting future states as video, converting those predictions into actions, and then re-observing the world in a closed loop. Unlike traditional methods relying on teleoperated robot trajectories, Rhoda pre-trains its models on vast amounts of internet video – hundreds of millions of examples – to build a strong understanding of motion, physics, and physical interactions. This is supplemented with smaller datasets of robot-specific data to fine-tune behaviors. CEO Jagdeep Singh, also founder of QuantumScape and Infinera, noted that this approach allows the robots to generalize better than those trained via teleoperation. The $450 million Series A funding will fuel research and engineering, expand industrial deployments, and grow Rhoda's team. Key investors include Capricorn Investment Group, Khosla Ventures, Temasek, and John Doerr. Rhoda is working with leading industrial partners in manufacturing and logistics, already demonstrating autonomous operation in production settings. Rhoda joins a growing field of "Physical AI" companies, where AI bridges the gap between the digital and physical, enabling machines to perceive, reason, and interact with 3D environments in real-time. Other companies in this space include Figure AI, Skild AI and 1X Technologies. This differs from traditional AI, which primarily "thinks", while Physical AI "thinks, then acts".
