Credit: AgiBot
TMTPOST -- Shanghai-based robotics start-up AgiBot has unveiled Genie Envisioner (GE), a real-world-oriented, unified video-generating platform that integrates prediction, policy learning, and neural simulation — a first-of-its-kind system in the industry, the company announced on Thursday.
According to AgiBot, the platform represents a major leap in the development of general-purpose, instruction-driven embodied intelligence. “World models for robotics should learn, act, and evaluate in one loop. We’re releasing Genie Envisioner: a unified, video-generative platform that integrates prediction, policy learning, and neural simulation together,” the company said on social media platform X last week.
Unlike conventional robotic training systems, which typically separate data collection, training, and evaluation into distinct stages, GE consolidates these processes into a single cohesive framework. The platform is designed to accelerate learning and improve real-world performance by allowing robots to visualize, plan, and execute tasks within the same loop.
At the heart of the platform is GE-Base, a large-scale, instruction-driven video diffusion model that captures the spatial, temporal, and semantic dynamics of real-world tasks. Trained on roughly 3,000 hours of video paired with language instructions — encompassing more than 1 million real-world robotic manipulation episodes — GE-Base establishes a detailed mapping from language commands to an embodied visual space. This enables robots to interpret complex instructions and translate them into coordinated physical actions.
The company says its vision-centric world modeling approach marks a shift from passive execution to an active “imagine-verify-act” paradigm. In practice, this means robots can simulate possible outcomes, verify predicted results, and then act accordingly — a method expected to significantly enhance adaptability and precision in dynamic environments.
AgiBot plans to make all code, models, and benchmarks related to GE open source, a move it says will foster broader adoption and innovation in the robotics community. Future development will include expanding sensor modalities to support full-body mobility and human-robot collaboration, with the goal of driving advancements in intelligent manufacturing and service robots.
The platform’s capabilities have already been demonstrated in extensive real-world testing. Robots powered by GE have successfully handled tasks such as folding clothes, sorting items on conveyor belts, making sandwiches, pouring tea, and operating microwave ovens. At the recent World Robot Conference in Beijing, these robots achieved a success rate that exceeded industry averages in complex task execution.
AgiBot emphasizes that GE’s high-fidelity physical simulation allows robots not only to understand their current environment but also to predict how it will change during interaction — a capability that could prove vital in settings ranging from factory floors to household environments.
Zhong Xiangyun, a humanoid robot industry observer, said the launch of GE reflects China’s growing strength in cutting-edge robotics research. “This is more than a new product — it’s a foundational shift in how robots can be trained and deployed. It opens the door to more efficient, scalable, and intelligent robot systems,” Zhong noted.
With the rollout of Genie Envisioner, AgiBot has positioned itself at the forefront of a new era in robotics — one where machines not only follow instructions but actively understand, adapt to, and interact with the complex physical world around them.
