Genie Envisioner - An open-source platform for robot world models launched by Zhiyuan.

What is Genie Envisioner?

Genie Envisioner is the first open-source robot world model platform launched by LogicNet. The platform integrates policy learning, evaluation, and simulation functions through a unified video generation framework. Core components include GE-Base (a large-scale instruction-conditional video diffusion model), GE-Act (a motion trajectory decoder), GE-Sim (a neural simulator), and EWMBench (a standardized benchmark suite). The platform supports policy generalization across robot forms, enabling robots to achieve precise operations in complex tasks, promoting the development of embodied intelligence, and providing strong support for robotics research and applications.

Genie Envisioner's main functions

• Strategy learningBased on GE-Base, the system captures the dynamics of the robot's interaction with the environment and generates strategies for action decision-making.
• Action generationIt maps the representation of the latent space to executable motion trajectories, supporting multiple robot forms.
• Simulation supportIt provides a high-fidelity simulation environment for closed-loop testing and optimization of strategies.
• Performance evaluationProvides standardized benchmarks to measure visual fidelity, physical consistency, and command-action alignment.

The technical principles of Genie Envisioner

• GE-BaseGE-Base is a large-scale command-conditional video diffusion model that captures the spatial, temporal, and semantic dynamics of robot interactions. It represents complex robot interactions in a structured latent space, facilitating subsequent processing.
• GE-ActGE-Act uses a lightweight stream matching decoder to map representations of the latent space to executable motion trajectories. It supports policy transfer between various robot forms with minimal supervision signals.
• GE-SimGE-Sim is a motion-conditional neural simulator used to generate high-fidelity playbacks. It supports strategy development and optimization in a virtual environment, reducing the need for physical experiments.
• EWMBenchEWMBench provides a standardized test suite for measuring a model's visual fidelity, physical consistency, and instruction-action alignment. This helps researchers and developers evaluate and optimize model performance.

Genie Envisioner's project address

• Project official websitehttps://genie-envisioner.github.io/
• GitHub repositoryhttps://github.com/AgibotTech/Genie-Envisioner
• arXiv technical paper: https://arxiv.org/pdf/2508.05635

Application scenarios of Genie Envisioner

• Industrial AutomationOn factory production lines, it helps robots complete complex assembly, handling, and quality inspection tasks more accurately, thereby improving production efficiency and product quality.
• Logistics and WarehousingUsed in logistics centers for sorting and handling goods, robots can quickly identify and process items of different shapes and sizes according to instructions, thus optimizing the logistics process.
• service robotsIn restaurant, hotel, or home environments, robots can be given the ability to understand and execute human instructions, providing smarter services such as food delivery, cleaning, and item delivery.
• Medical assistanceIn medical settings, robots assist in surgery, provide rehabilitation training, or deliver medications, improving the accuracy and efficiency of medical services.
• Education and ResearchIt provides a powerful experimental platform for universities and research institutions, supports research in robot learning, artificial intelligence and embodied intelligence, and promotes the development of related technologies.