Zhiyuan Robotics (AgiBot) has announced the release of Genie Envisioner (GE), a groundbreaking open-source platform for real-world robot control through unified world modeling. The system integrates future frame prediction, policy learning, and simulation evaluation into a single closed-loop architecture centered on video generation.
The platform was trained on approximately 3,000 hours of real robot operation video data, giving it exceptional capabilities in cross-platform generalization and long-term task execution. GE employs a vision-centered modeling paradigm that differs from mainstream VLA methods, preserving spatial structure and temporal evolution information.
The GE platform consists of three integrated components: GE-Base (autoregressive video generation with multi-view capabilities), GE-Act (plug-and-play action module), and GE-Sim (neural simulator for policy evaluation). The team also developed EWMBench, a standardized benchmark suite for world model quality assessment. Zhiyuan plans to open-source all code, models, and evaluation tools.
