Zhiyuan Robotics has announced the launch of Genie Studio, the industry's first one-stop development platform specifically designed for embodied AI scenarios. Genie Studio boasts full-chain product capabilities encompassing data acquisition, model training, simulation evaluation, and model inference, providing developers with a standardized solution from "acquisition" to "training" to "testing" and finally to "inference." This significantly lowers the development threshold and improves efficiency.
Genie Studio's data acquisition module has fundamentally redesigned the robot data lifecycle, creating a one-stop data hub that supports multi-body and multi-end device management and provides full-process functional services. Through the reconstruction of simulated scenarios and multiple expert trajectory generation strategies, combined with domain randomization and data augmentation techniques, the platform supports automated acquisition of large-scale synthetic data. For complex operational tasks, the platform allows developers to collect fine-grained operational data through teleoperation and combines trajectory auto-generalization to achieve large-scale data augmentation.
In terms of model training, Genie Studio, as a new generation of embodied AI training engine, has achieved a comprehensive upgrade in R&D experience. The platform features a unique video-based data loading scheme, supports seamless integration of various open-source datasets, deeply integrates base models and various open-source end-to-end embodied models, provides pre-training/fine-tuning task templates, and, coupled with its self-developed training and R&D framework, continuously improves training efficiency. Through the deep integration of standardized resource pool management, training images, model management, and compilation and deployment pipelines, the platform achieves end-to-end rapid model delivery.
The simulation evaluation module is based on a large-scale, high-precision 3D asset library, providing developers with a wealth of simulation scenarios, object assets, and robot models. It can automatically generate diverse scenarios and tasks, accurately reproducing various robot operating environments. The platform builds a complete simulation evaluation process from scene generation, model inference, and benchmark testing to result visualization. Through an efficient simulation toolchain, it helps users quickly verify algorithm performance and optimize models.
The model inference module directly addresses industry pain points, reconstructing the deployment paradigm through four core technologies: Model compilation technology breaks through hardware heterogeneity barriers, achieving "develop once, deploy everywhere" for algorithms; The GDK (Genie Development Kit) development components use standardized interfaces to connect to the robot hardware layer, building a stable and reliable real-machine communication base; The cloud-integrated inference framework simplifies the model deployment process to "one-click real-machine deployment," facilitating the migration of algorithm development from cloud environments to physical robots; Coupled with a full-chain evaluation system, it achieves quantitative evaluation of deployment effects and establishes a closed-loop system of application version management, remote diagnosis, and dynamic optimization.
The launch of Genie Studio marks a new stage in embodied AI development. The platform not only provides developers with powerful tools but also promotes the transformation of embodied AI from code executor to environment perceiver, accelerating the "last mile" of large-scale embodied AI implementation. Through Genie Studio, developers can more efficiently perform data acquisition, model training, simulation evaluation, and model inference, driving the rapid development and application of embodied AI technology.
Official Website
https://genie.agibot.com/geniestudio
