Search papers, labs, and topics across Lattice.
GeniWorld is an interactive world model designed for robotic manipulation that leverages pretrained video generative models to enhance generalization across unseen environments. By decoupling embodiment kinematics from environmental dynamics and employing URDF-based rendering for visual action representations, the model achieves robust zero-shot generalization and superior performance even with limited training data. The integration of an autoregressive video prediction model with high-frequency kinematic control allows for effective closed-loop interactions, making GeniWorld a scalable and reliable policy evaluator in complex scenarios.
GeniWorld achieves robust zero-shot generalization in robotic manipulation, outperforming traditional models even with minimal training data.
Generalist robot policies exhibit strong capabilities, but their robustness in complex and unseen environments remains limited. Scaling robot learning and evaluation in diverse real-world environments remains costly and challenging. Action-conditioned world models offer a promising alternative, but they often suffer from limited action controllability and poor generalization to out-of-distribution (OOD) scenarios. To this end, we present GeniWorld, an interactive world model for robots that generalizes robustly across unseen scenarios. Building on pretrained video generative models, we use URDF-based rendering to transform numerical actions into visual action representations, enabling spatially grounded action control. By explicitly decoupling embodiment kinematics from environmental dynamics, our model mitigates scene overfitting and facilitates modeling of robot-environment interactions. To achieve closed-loop control, we construct an autoregressive video prediction model integrated with high-frequency robot kinematic control, enabling interaction with both robot policies and human teleoperators. In our experiments, even when trained solely on limited fixed-scene data, our model achieves superior in-domain performance and robust zero-shot generalization to highly randomized, unseen environments. For downstream applications, GeniWorld serves as a scalable policy evaluator that remains reliable under environmental perturbations. Furthermore, even with limited real-world demonstrations, GeniWorld generates diverse manipulation trajectories within the world model, improving downstream policy performance and robustness in complex environments.
Tsinghua AI