HiDream.ai launches embodied world model HiDream-O1-Embodied, tops RoboColiseum robustness leaderboard

HiDream.ai has released HiDream-O1-Embodied, an embodied world model built on a native full-modal architecture that unifies image, video, 3D and action signals into a single representation framework to enhance robotic physical perception and dynamic prediction capabilities, the company said, marking a strategic step toward building a unified world model foundation that bridges understanding, reasoning and execution across modalities; concurrently, the model made its debut on the embodied AI evaluation platform RoboColiseum – a simulation-based benchmark featuring 78 high-fidelity tasks across instruction following, spatial understanding, robustness and general manipulation – and topped the Robustness sub-leaderboard with an average score of 0.692, a category widely regarded as the most challenging for testing stability and generalisation under non-ideal conditions such as altered lighting, camera shifts, material changes and varied instructions.